Liouville Theory and the Virasoro Bootstrap

Liouville theory is the canonical exactly solvable CFT with a continuous Virasoro spectrum. Its bootstrap data combine a reflection identification, delta-normalized states, special-function three-point constants, and Virasoro blocks integrated along a prescribed contour. The formulas below describe spacelike Liouville theory with real $b>0$ and generic external momenta; analytic continuation outside that domain requires an explicit pole prescription.

Evidence cutoff. This account reflects the sources consulted through 2026-08-09. Statements about analytic continuation and crossing are conditional on the spectrum, meromorphy, contour, and normalization hypotheses stated below; they are not claims about every theory called “Liouville” or every continuation to $c\leq1$.

Required background. Noncompact CFTs and continuous spectra provide direct-integral normalization and contour language. Highest-weight modules, null states, and the Kac determinant provide the degenerate Virasoro representations used to obtain shift equations.

Helpful background. Partial waves and the shadow formalism give a higher-dimensional comparison for continuum decompositions and normalization choices.

Action, central charge, and spectrum

On a Euclidean surface with reference metric $g$, take

$$S_L[\phi] =\frac{1}{4\pi}\int d^2x\sqrt g\, \left[g^{ab}\partial_a\phi\partial_b\phi +Q R_g\phi+4\pi\mu e^{2b\phi}\right],$$

with

$$Q=b+b^{-1}, \qquad c_L=1+6Q^2, \qquad \mu>0.$$

The exponential primary

$$V_\alpha(z,\bar z)=e^{2\alpha\phi(z,\bar z)}$$

has weights

$$h_\alpha=\bar h_\alpha=\alpha(Q-\alpha).$$

Normalizable scattering states lie on

$$\alpha=\frac Q2+iP, \qquad P\in\mathbb R, \qquad h_P=\frac{Q^2}{4}+P^2.$$

The reflection relation

$$V_\alpha=R(\alpha)V_{Q-\alpha}, \qquad R(\alpha)R(Q-\alpha)=1,$$

identifies $P$ and $-P$. We therefore choose a reflected basis $\widehat V_P$ with $P\geq0$ and declare

$$\langle P|P'\rangle=2\pi\delta(P-P'), \qquad \mathbf1=\int_0^\infty\frac{dP}{2\pi}|P\rangle\langle P|.$$

The hat absorbs the reflection amplitude and a normalization factor. Raw $V_\alpha$ correlators instead contain both $\delta(P-P')$ and the reflected $\delta(P+P')$; mixing these two conventions double-counts the continuum. The spectrum, two-point normalization, and degenerate analytic continuations are developed in Ribault 2018, §§3.1.1–3.1.2, pp. 69–73.

The interacting zero mode

On the sphere, write $\phi=\phi_0+\widetilde\phi$ and let

$$\mathcal A[\widetilde\phi] =\int d^2x\sqrt g\,e^{2b\widetilde\phi}.$$

The constant-mode part of an $n$-point path integral is proportional to

$$\int_{-\infty}^{\infty}d\phi_0\, \exp\!\left[ 2\left(\sum_i\alpha_i-Q\right)\phi_0 -\mu\mathcal A e^{2b\phi_0} \right].$$

For $\operatorname{Re}(\sum_i\alpha_i-Q)>0$, the substitution $y=\mu\mathcal A e^{2b\phi_0}$ gives

$$\frac{1}{2b} (\mu\mathcal A)^{-s}\Gamma(s), \qquad s=\frac{\sum_i\alpha_i-Q}{b}.$$

Outside this convergence half-plane, the expression is defined by meromorphic continuation rather than by the original absolutely convergent zero-mode integral. Its gamma-function poles foreshadow the discrete residues that appear when an OPE contour is continued. This is an interacting analogue of the free-boson zero-mode issue, but it does not produce an ordinary momentum-conservation delta function because the exponential potential breaks translations of $\phi$.

The DOZZ three-point constant

Let

$$\gamma(x)=\frac{\Gamma(x)}{\Gamma(1-x)}, \qquad A=\pi\mu\gamma(b^2)b^{2-2b^2}.$$

In the standard exponential-field normalization, the Dorn–Otto–Zamolodchikov–Zamolodchikov three-point constant is

$$\begin{aligned} C(\alpha_1,\alpha_2,\alpha_3) ={}&A^{(Q-\alpha_1-\alpha_2-\alpha_3)/b}\\ &\times \frac{\Upsilon_b'(0) \prod_{i=1}^3\Upsilon_b(2\alpha_i)} {\Upsilon_b(\alpha_1+\alpha_2+\alpha_3-Q) \Upsilon_b(\alpha_1+\alpha_2-\alpha_3)}\\ &\times \frac{1} {\Upsilon_b(\alpha_1+\alpha_3-\alpha_2) \Upsilon_b(\alpha_2+\alpha_3-\alpha_1)}. \end{aligned}$$

The entire function $\Upsilon_b$ is fixed, up to normalization, by its zero set, the duality $b\leftrightarrow b^{-1}$, and the shift equations

$$\Upsilon_b(x+b) =\gamma(bx)b^{1-2bx}\Upsilon_b(x),$$ $$\Upsilon_b(x+b^{-1}) =\gamma(x/b)b^{2x/b-1}\Upsilon_b(x).$$

The original derivation, reflection amplitude, and four-point construction appear in Zamolodchikov and Zamolodchikov 1996, §§2–5, pp. 579–592. The formula is meromorphic in the $\alpha_i$; its zeros and poles are part of the contour data, not removable numerical nuisances.

Degenerate fields and the shift equation

The degenerate field $V_{-b/2}$ is not a normalizable state on the principal-series contour; it is introduced as a meromorphic continuation of correlation functions. It has a level-two null descendant,

$$\left(L_{-1}^2+b^2L_{-2}\right)V_{-b/2}=0,$$

and its OPE with a generic exponential contains only two Virasoro families:

$$V_{-b/2}V_\alpha \sim C_+(\alpha)V_{\alpha-b/2} +C_-(\alpha)V_{\alpha+b/2}.$$

The null relation turns a four-point Ward identity into a second-order BPZ equation. Comparing its two finite channel decompositions relates the coefficients with intermediate momenta shifted by $\pm b/2$. After eliminating the degenerate OPE normalization, one obtains a $b$-shift equation for the generic three-point constant; using the dual field $V_{-1/(2b)}$ gives the $b^{-1}$ shift. This derivation and the normalization-independent form of the shift equations are given in Ribault 2018, §3.1.2, pp. 71–75.

The DOZZ formula passes this test directly. Define

$$\begin{gathered} x_0=\alpha_1+\alpha_2+\alpha_3-Q, \qquad x_1=\alpha_1+\alpha_2-\alpha_3,\\ x_2=\alpha_1+\alpha_3-\alpha_2, \qquad x_3=\alpha_2+\alpha_3-\alpha_1. \end{gathered}$$

Taking the ratio at $\alpha_1+b$ and $\alpha_1$ gives

$$\begin{aligned} \frac{C(\alpha_1+b,\alpha_2,\alpha_3)} {C(\alpha_1,\alpha_2,\alpha_3)} ={}&A^{-1} \frac{\Upsilon_b(2\alpha_1+2b)} {\Upsilon_b(2\alpha_1)}\\ &\times \frac{\Upsilon_b(x_0)\Upsilon_b(x_1)} {\Upsilon_b(x_0+b)\Upsilon_b(x_1+b)}\\ &\times \frac{\Upsilon_b(x_2)\Upsilon_b(x_3)} {\Upsilon_b(x_2+b)\Upsilon_b(x_3-b)}. \end{aligned}$$

Applying the $\Upsilon_b$ shift identity converts this expression into the gamma-function BPZ shift relation. The dual identity supplies the independent $b^{-1}$ shift. The two shifts, reflection symmetry, meromorphy, and growth assumptions remove periodic ambiguities for generic irrational $b^2$; omitting those assumptions overstates what the finite BPZ equation alone proves.

Four-point crossing as a contour integral

Place four operators at $0,z,1,\infty$. In the reflected $P\geq0$ normalization declared above, an $s$-channel decomposition has the form

$$\mathcal G(z,\bar z) =\int_0^\infty\frac{dP}{2\pi}\, \mathcal C_{12}(P)\mathcal C_{P34} \left|\mathcal F_P^{(s)}(z)\right|^2.$$

The coefficients $\mathcal C$ are the DOZZ constants multiplied by the conversion factors from raw $V_\alpha$ fields to $\widehat V_P$ states. In the original full-line convention, the equivalent formula is one half of an integral over $P\in\mathbb R$. The choice must be made once and used in the two-point function, completeness relation, and four-point measure.

For real external $\alpha_i$ in the strip where no DOZZ pole crosses $\alpha=Q/2+i\mathbb R$, the contour is literal. When the $\alpha_i$ are analytically continued, moving poles can cross it. The continued correlator is then

$$\mathcal G =\int_{\mathcal C_{\mathrm{continued}}}d\mu(P)\, \rho(P)|\mathcal F_P|^2 +2\pi i\sum_{P_*} \operatorname*{Res}_{P=P_*} \bigl[d\mu(P)\rho(P)|\mathcal F_P|^2\bigr],$$

with signs fixed by the contour orientation. Zamolodchikov and Zamolodchikov 1996, §2, pp. 581–583 state the initial domain and the need for additional discrete terms after pole crossings. Crossing means that the resulting analytically continued function agrees with the corresponding $t$-channel construction, including the same residues—not that the unmodified real-$P$ integral works for every external momentum.

Independent checks and limitations

• Reflection: verify that integrating over $P\geq0$ gives the same result as the half-weighted full-line integral. A factor-of-two discrepancy signals double counting.
• Zero mode: reproduce the power of $\mu$ and the gamma-function poles from the constant-mode integral.
• Degenerate limit: let one external momentum approach $-b/2$; the continuum contour must pinch or reduce to the two allowed BPZ channels with the correct residues.
• Duality: repeat the shift test with $b^{-1}$. A proposed three-point constant that satisfies only one shift equation can retain an undetected periodic factor.
• Pole crossing: vary an external $\alpha_i$ continuously and record each pole crossing before comparing channels.
• Block normalization: check the leading OPE behavior $\mathcal F_P^{(s)}(z)\sim z^{h_P-h_1-h_2}(1+O(z))$ in the same convention used by the structure constants.
• Numerics: if the $q$-series for a Virasoro block is truncated, report the radial domain, truncation order, precision, and tail estimate. Numerical agreement supports a specified integral; it does not remove analytic-continuation hypotheses.

Spacelike $b>0$ gives $c_L\geq25$. Timelike or $c\leq1$ continuations involve different contours, spectra, and sometimes different structure constants; the formulas on this page cannot be transferred to them merely by substituting an imaginary $b$.

The channel-changing transform of the Virasoro blocks themselves is the subject of Irrational CFT, fusion kernels, and crossing.

Exercises

Check the Liouville continuum weight

Show that $h_\alpha$ is real and bounded below on $\alpha=Q/2+iP$ with real $P$.

Solution

$$h_\alpha =\left(\frac Q2+iP\right) \left(\frac Q2-iP\right) =\frac{Q^2}{4}+P^2.$$

It is real and has minimum $Q^2/4$ at $P=0$. The states remain delta-normalized because $P$ is continuous.

Evaluate the zero-mode integral

For $\operatorname{Re}s>0$, evaluate

$$I=\int_{-\infty}^{\infty}d\phi_0\, e^{2bs\phi_0-\mu\mathcal A e^{2b\phi_0}}.$$

Solution

Set $y=\mu\mathcal A e^{2b\phi_0}$, so $d\phi_0=dy/(2by)$ and $e^{2bs\phi_0}=(y/\mu\mathcal A)^s$. Then

$$I=\frac{(\mu\mathcal A)^{-s}}{2b} \int_0^\infty dy\,y^{s-1}e^{-y} =\frac{(\mu\mathcal A)^{-s}}{2b}\Gamma(s).$$

The poles after meromorphic continuation occur at $s=0,-1,-2,\ldots$.

References

• Ribault, Sylvain. “Conformal Field Theory on the Plane.” SciPost Physics Lecture Notes 1 (2018). DOI.
• Zamolodchikov, Alexander B., and Al. B. Zamolodchikov. “Structure Constants and Conformal Bootstrap in Liouville Field Theory.” Nuclear Physics B 477 (1996): 577–605. DOI.