Conformal Bootstrap Principle

• Conformal Bootstrap Principle is a framework that uses conformal symmetry, the operator product expansion, and crossing relations to rigorously constrain and solve CFTs without relying on a microscopic Lagrangian.
• It employs both analytic techniques, as seen in 2D minimal models and Liouville theory, and numerical methods to carve out precise bounds or islands for operator dimensions and coupling constants.
• The approach underpins high-precision determinations in models like the 3D Ising universality class, illustrating its impact on understanding critical phenomena and scaling behaviors.

The conformal bootstrap principle is the program of solving or rigorously constraining a conformal field theory by using only conformal symmetry, the operator product expansion, crossing symmetry, and unitarity, without relying on a microscopic Lagrangian. In this framework, the essential dynamical data are the spectrum of local primary operators—specified by their scaling dimensions and spins—and the OPE coefficients that govern three-point couplings. Conformal invariance fixes kinematics, while associativity of the OPE turns four-point functions into nontrivial consistency conditions; in two dimensions these conditions can be exact and analytic, whereas in higher dimensions they are often implemented numerically to obtain rigorous bounds or isolated “islands” of admissible CFT data (Simmons-Duffin, 2016, Ribault, 2014, Rychkov, 2 Sep 2025).

1. Historical development

The modern formulation has roots in the 1960s and 1970s, when scale invariance and anomalous dimensions emerged in the study of strong interactions and critical phenomena. Polyakov argued for conformal invariance at criticality and formulated a non-Hamiltonian, crossing-based program. The Rome group developed OPE methods, conformal partial waves, shadow formalism, and early positivity bounds, while Mack and the Sofia group pursued harmonic analysis and wrote nonperturbative crossing equations and classification results for unitary representations. This “old bootstrap” was later eclipsed by Wilson–Fisher renormalization-group methods, but its basic philosophy already matched the contemporary viewpoint: consistency conditions and symmetry should determine the theory (Rychkov, 2 Sep 2025).

The first decisive exact realizations came in two dimensions. The BPZ revolution showed that Virasoro symmetry, degenerate representations, and crossing symmetry solve minimal models exactly; later developments extended the analytic bootstrap to Liouville theory, generalized minimal models, free bosons, and WZW models. In higher dimensions, the numerical bootstrap re-emerged in the 2000s, especially after the linear-functional and positivity methods introduced rigorous bounds from crossing around the symmetric point. The subsequent discovery of kinks and then mixed-correlator “islands” for the 3D Ising universality class established the bootstrap as a quantitative nonperturbative method (Ribault, 2014, Simmons-Duffin, 2016).

2. CFT data and consistency conditions

A conformal field theory is specified by its spectrum of primary operators and by OPE coefficients. In the standard scalar notation,

$$\phi_i(x)\,\phi_j(0)=\sum_{\mathcal O} C_{ij\mathcal O}\,|x|^{\Delta_{\mathcal O}-\Delta_i-\Delta_j}\,\mathcal O(0)+\cdots.$$

The sum runs over primaries and descendants, and conformal covariance fixes the descendant structure once the primaries and three-point data are known (Rychkov, 2 Sep 2025, Simmons-Duffin, 2016).

Several structural ingredients are indispensable. Conformal invariance fixes the form of two- and three-point functions and constrains higher-point correlators to depend only on conformal cross-ratios. Radial quantization and the state–operator correspondence identify local operators with states on spheres, making the OPE an operator/state expansion. Reflection positivity implies positivity of norms and reality of appropriate OPE coefficients in unitary theories. In $d$ dimensions, the standard unitarity bounds are

$$\Delta \ge \frac{d-2}{2}\quad \text{for scalars},\qquad \Delta \ge \ell+d-2\quad \text{for spin }\ell>0.$$

Operators saturating these bounds include conserved currents and the stress tensor (Simmons-Duffin, 2016, Rychkov, 2020).

The stress tensor is central to the bootstrap principle. In a local unitary CFT there must be a conserved symmetric traceless tensor $T_{\mu\nu}$ of dimension $\Delta_T=d$, and Ward identities fix the normalization of certain three-point structures involving $T_{\mu\nu}$. Global symmetries further decompose the operator algebra into sectors with selection rules. In the 3D Ising universality class, for example, the $\mathbb Z_2$ symmetry splits operators into even and odd sectors, so that $\sigma\times\sigma$ is $\mathbb Z_2$-even, while $\sigma\times\varepsilon$ is $d$0-odd (Rychkov, 2020).

3. Four-point functions, conformal blocks, and crossing symmetry

For identical scalar primaries $d$1 of dimension $d$2, conformal invariance implies

$d$3

with conformal cross-ratios

$d$4

The reduced correlator admits a conformal block decomposition

$d$5

where $d$6 are universal kinematic functions determined by symmetry and spacetime dimension (Rychkov, 2 Sep 2025, Simmons-Duffin, 2016).

Crossing symmetry is the statement that different OPE channel decompositions of the same four-point function must agree. For identical scalars,

$d$7

Equivalently,

$d$8

This is the core bootstrap equation: a functional equation in two variables, with positivity when the external operators are identical and the theory is unitary (Simmons-Duffin, 2016, Rychkov, 2020).

In two dimensions the structure becomes much stronger. Local conformal transformations generate the Virasoro algebra,

$d$9

with an independent antiholomorphic copy. Four-point functions factorize into holomorphic and antiholomorphic Virasoro blocks,

$$\Delta \ge \frac{d-2}{2}\quad \text{for scalars},\qquad \Delta \ge \ell+d-2\quad \text{for spin }\ell>0.$$0

and crossing equates this decomposition to the one obtained under $$\Delta \ge \frac{d-2}{2}\quad \text{for scalars},\qquad \Delta \ge \ell+d-2\quad \text{for spin }\ell>0.$$1. Degenerate fields then satisfy BPZ differential equations, reducing the bootstrap problem to special-function theory in many solvable models (Ribault, 2014).

4. Numerical bootstrap and the emergence of islands

In higher dimensions, exact analytic solutions are rare, so crossing is usually cast as an optimization problem. The standard numerical method chooses linear functionals built from derivatives at the Euclidean crossing-symmetric point. In one common basis,

$$\Delta \ge \frac{d-2}{2}\quad \text{for scalars},\qquad \Delta \ge \ell+d-2\quad \text{for spin }\ell>0.$$2

or, equivalently in some formulations,

$$\Delta \ge \frac{d-2}{2}\quad \text{for scalars},\qquad \Delta \ge \ell+d-2\quad \text{for spin }\ell>0.$$3

If a functional can be found that is nonnegative on all admissible block contributions but negative on the crossing target, the assumed spectrum is excluded. Increasing the derivative order $$\Delta \ge \frac{d-2}{2}\quad \text{for scalars},\qquad \Delta \ge \ell+d-2\quad \text{for spin }\ell>0.$$4 strengthens the exclusion (Simmons-Duffin, 2016, Rychkov, 2020).

For multiple correlators, positivity becomes matrix-valued. In the mixed-correlator 3D Ising bootstrap, one studies $$\Delta \ge \frac{d-2}{2}\quad \text{for scalars},\qquad \Delta \ge \ell+d-2\quad \text{for spin }\ell>0.$$5, $$\Delta \ge \frac{d-2}{2}\quad \text{for scalars},\qquad \Delta \ge \ell+d-2\quad \text{for spin }\ell>0.$$6, and $$\Delta \ge \frac{d-2}{2}\quad \text{for scalars},\qquad \Delta \ge \ell+d-2\quad \text{for spin }\ell>0.$$7 simultaneously. Products such as $$\Delta \ge \frac{d-2}{2}\quad \text{for scalars},\qquad \Delta \ge \ell+d-2\quad \text{for spin }\ell>0.$$8 are organized into positive semidefinite matrices,

$$\Delta \ge \frac{d-2}{2}\quad \text{for scalars},\qquad \Delta \ge \ell+d-2\quad \text{for spin }\ell>0.$$9

and the resulting feasibility problem is solved as a semidefinite program using SDPB (Rychkov, 2020).

The canonical case study is the 3D Ising CFT. Assuming $T_{\mu\nu}$0 symmetry, a $T_{\mu\nu}$1-odd scalar $T_{\mu\nu}$2, a $T_{\mu\nu}$3-even scalar $T_{\mu\nu}$4, and that these are the only relevant scalars, the mixed-correlator bootstrap carves out a tiny allowed region in the $T_{\mu\nu}$5-plane. The resulting island lies near

$T_{\mu\nu}$6

which implies

$T_{\mu\nu}$7

and numerically yields

$T_{\mu\nu}$8

These values agree extremely well with Monte Carlo and high-order RG analyses, and the bootstrap determinations are described as currently the most precise in the cited review (Rychkov, 2020). A later historical survey reports that, with stress-tensor correlators included, the leading 3D Ising exponents reach accuracy at the $T_{\mu\nu}$9–$\Delta_T=d$0 level (Rychkov, 2 Sep 2025).

5. Analytic bootstrap in two dimensions and beyond

The analytic bootstrap is most developed in two dimensions because Virasoro symmetry factorizes correlators into holomorphic and antiholomorphic sectors and greatly enlarges the set of exact constraints. Minimal models are solved by combining degenerate-field BPZ equations, fusion rules, and crossing symmetry. Generalized minimal models extend this logic to infinite discrete spectra, while Liouville theory realizes a non-rational CFT with continuous spectrum and exact three-point structure constants given by the DOZZ formula (Ribault, 2014).

A rigorous realization of the bootstrap principle in Liouville theory was established by proving that the probabilistic construction of the theory reproduces the conformal block integral representation of four-point functions and satisfies crossing symmetry. In that setting, the spectrum is continuous,

$\Delta_T=d$1

and the four-point function takes the DOZZ-weighted conformal block form

$\Delta_T=d$2

with equality of the $\Delta_T=d$3- and $\Delta_T=d$4-channel representations providing the crossing relation (Guillarmou et al., 2020).

Beyond two dimensions, several analytic variants pursue the same principle in different bases. The Polyakov–Mellin bootstrap replaces single-channel conformal blocks by crossing-symmetric exchange Witten diagrams in Mellin space and demands cancellation of spurious poles. This produces explicit infinite families of constraints on dimensions and OPE coefficients and has been used to recover Wilson–Fisher $\Delta_T=d$5-expansion data and higher-order OPE coefficients (Gopakumar et al., 2016). Other approaches include determinant methods based on small minors and truncated crossing, which can access non-unitary theories such as Yang–Lee without positivity assumptions (Gliozzi, 2013).

6. Scope, variants, limitations, and open problems

A common misconception is that the bootstrap is only a boundary-finding technology. That is accurate for positivity-based exclusion methods in their standard form, but not for the broader family of bootstrap techniques. The determinant method can access theories in the interior of the allowed region and non-unitary theories when positivity is absent (Gliozzi, 2013). A Monte Carlo approach based on minimizing a weighted residual of truncated crossing equations searches directly for approximate solutions, including solutions away from extremality, and was shown to recover the 2D and 3D Ising theories as well as the 2D Yang–Lee model (Laio et al., 2022). Multipoint or “effective” bootstrap methods sample the cross-ratio plane away from the symmetric point and use explicit tail bounds to integrate out heavy operators (Echeverri et al., 2016).

The method also has structural limitations. It relies on the CFT axioms and, in statistical-mechanical applications, on the assumption that critical points are described by local unitary CFTs. In $\Delta_T=d$6, rigorous derivations of this assumption are described in the Ising review as future work. Numerical results depend on finite truncations, derivative order $\Delta_T=d$7, machine precision, and spectral gap assumptions, although robustness is typically tested by varying these inputs. Even when the allowed region shrinks dramatically, the outcome is still an allowed set rather than a constructive definition of the theory (Rychkov, 2020).

Current research directions include proving uniqueness of the 3D Ising CFT under its standard assumptions, deriving no-go results for putative second-order transitions, improving convergence for large external dimensions through analytic functional bases, and extending mixed-correlator bootstrap to spinning operators, stress-tensor sectors, and higher-point constraints (Rychkov, 2 Sep 2025). These developments suggest that the bootstrap principle is not a single algorithm but a general nonperturbative framework: conformal symmetry fixes kinematics, the OPE organizes local dynamics, associativity becomes crossing symmetry, and unitarity or other structural conditions restrict the admissible solutions.