Form-Factor Bootstrap and Correlator Expansions

An exact two-body S matrix determines how multiparticle matrix elements of a local operator are permuted and analytically continued, but it does not determine which operator is being studied or prove convergence of its correlator expansion. The form-factor bootstrap combines Watson exchange, crossing or cyclicity, kinematic and bound-state residues, Lorentz covariance, operator quantum numbers, and normalization. Correlators then follow from a spectral sum whose omitted tail must be tested at the quoted separation.

Required background. Exact S-matrix bootstrap, CDD freedom, and completeness supplies the scattering data and analytic conventions. Spectral decomposition of two-point functions supplies completeness and spectral positivity. Helpful background. Bound-state poles and fusion supplies bound-state residue conventions, while form factors and local-operator insertions supplies the general LSZ meaning of an operator matrix element.

Form factors and state normalization

Use asymptotic states normalized by

$$\langle a,\theta'\mid b,\theta\rangle =2\pi\,\delta_{ab}\delta(\theta'-\theta).$$

For a local operator $\mathcal O$, define the incoming $n$-particle form factor

$$F_n^{\mathcal O\mid a_1\ldots a_n} (\theta_1,\ldots,\theta_n) = \langle0\lvert\mathcal O(0) \rvert a_1,\theta_1;\ldots;a_n,\theta_n\rangle_{\rm in}.$$

The rapidities are initially ordered on the real axis. Analytic continuation defines other orderings and crossed matrix elements. The operator label is essential: the same S matrix supports infinitely many local and semilocal operators with different spin, internal charge, normalization, ultraviolet dimension, and locality phase.

The foundational form-factor equations were derived from maximal analyticity and LSZ reasoning by Karowski and Weisz 1978, §§ 2–4, pp. 458–470. A concise convention-complete review is Babujian, Foerster, and Karowski 2006, §§ 2–3, pp. 2–6.

Watson exchange and cyclicity

For one diagonal species, exchanging adjacent rapidities gives Watson’s equation

$$\begin{aligned} F_n^{\mathcal O}(\ldots,\theta_i,\theta_{i+1},\ldots) ={}& S(\theta_i-\theta_{i+1}) \\ &\times F_n^{\mathcal O}(\ldots,\theta_{i+1},\theta_i,\ldots). \end{aligned}$$

Analytically taking the first particle around the operator gives the cyclic relation

$$F_n^{\mathcal O}(\theta_1+2\pi i,\theta_2,\ldots,\theta_n) = \ell_{\mathcal O}\, F_n^{\mathcal O}(\theta_2,\ldots,\theta_n,\theta_1).$$

Here $\ell_{\mathcal O}$ is the mutual-locality phase between the operator and the particle-creating field used to define the asymptotic state. It is one for a mutually local bosonic operator in this convention, but it need not be one for disorder, twist, or soliton fields. Dropping it changes the solution space.

Lorentz covariance adds, for an operator of spin $s_{\mathcal O}$,

$$F_n^{\mathcal O}(\theta_1+\lambda,\ldots,\theta_n+\lambda) =e^{s_{\mathcal O}\lambda} F_n^{\mathcal O}(\theta_1,\ldots,\theta_n),$$

where this equation defines the sign convention for $s_{\mathcal O}$.

Kinematic and bound-state residues

In the same scalar normalization, the annihilation or kinematic pole obeys

$$\begin{aligned} -i\,\underset{\theta'=\theta+i\pi}{\operatorname{Res}}\, F_{n+2}^{\mathcal O} (\theta',\theta,\theta_1,\ldots,\theta_n) ={}& \left[ 1-\ell_{\mathcal O} \prod_{j=1}^{n}S(\theta-\theta_j) \right] \\ &\times F_n^{\mathcal O}(\theta_1,\ldots,\theta_n). \end{aligned}$$

Other normalizations can carry charge-conjugation matrices, statistics factors, or a factor of two. The entire convention must be translated, not just the residue coefficient.

If particles $a$ and $b$ fuse to $c$ at $\theta_{ab}=iu_{ab}^c$, let $\Gamma_{ab}^c$ denote the fusion coefficient in the present form-factor state normalization. It is fixed from the S-matrix residue coupling $g_{ab}^c$ after the state-normalization and rapidity-Jacobian conversion; it is not an independent fit parameter. Choose its phase so that the bound-state residue is

$$-i\,\underset{\theta_{ab}=iu_{ab}^c}{\operatorname{Res}}\, F_{n+2}^{\mathcal O\mid ab a_1\ldots a_n} = \Gamma_{ab}^c\, F_{n+1}^{\mathcal O\mid c a_1\ldots a_n},$$

with the fused rapidity fixed by momentum conservation and with projector and phase factors restored in matrix channels. This equation tests consistency between the proposed particle spectrum and the operator matrix elements.

A minimal two-particle solution

Take the constant diagonal amplitude $S(\theta)=-1$ and a mutually local scalar operator, $\ell_{\mathcal O}=1$. Watson’s two-particle equation becomes

$$F_2(\theta)=-F_2(-\theta), \qquad \theta=\theta_1-\theta_2.$$

Cyclicity requires

$$F_2(\theta+2\pi i)=F_2(-\theta).$$

The function

$$F_{2,\min}(\theta)=C\sinh\!\left(\frac{\theta}{2}\right)$$

satisfies both equations:

$$F_{2,\min}(-\theta)=-F_{2,\min}(\theta), \qquad F_{2,\min}(\theta+2\pi i)=-F_{2,\min}(\theta).$$

The constant $C$ is not fixed by scattering. Operator normalization, short-distance behavior, internal symmetries, and residue equations select the physical form factor. Multiplying a minimal solution by suitable symmetric periodic functions can generate further solutions associated with other operators or descendants. Thus solving Watson’s equation is the start of operator identification, not its end.

Spectral reconstruction

For a Hermitian scalar operator at Euclidean separation $r>0$, insertion of asymptotic states gives the connected two-point function

$$\begin{aligned} C_{\mathcal O}(r) ={}& \sum_{n=1}^{\infty}\frac{1}{n!} \int\prod_{j=1}^{n}\frac{\mathrm d\theta_j}{2\pi}\, \left| F_n^{\mathcal O}(\theta_1,\ldots,\theta_n) \right|^2 \\ &\times \exp\!\left[ -r\sum_{j=1}^{n}m_{a_j}\cosh\theta_j \right], \end{aligned}$$

with an additional species sum when needed. The factorial removes overcounting in this unrestricted integration convention. At large $r$, the Boltzmann factor suppresses high particle number and rapidity, so a few sectors can be accurate. At short distance, many sectors can contribute and the expansion may converge slowly even though each retained form factor is exact.

For a positive spectral sum, define

$$C_{\le N}(r)=\sum_{n=1}^{N}C_n(r), \qquad \Delta_{N+1}(r)=C_{N+1}(r).$$

A practical tail test reports

$$\rho_N(r)= \frac{\Delta_{N+1}(r)} {C_{\le N+1}(r)}$$

over the full quoted range of $r$, repeats the last sector with a larger rapidity cutoff and finer quadrature, and compares with any ultraviolet sum rule or independent exact limit. The next sector is a convergence diagnostic, not a rigorous upper bound on the entire tail unless a monotone or geometric bound is proved. For nonpositive or mixed correlators, even the termwise positivity used in this diagnostic can fail.

Completeness at three levels

It is useful to distinguish:

1. S-matrix completeness: all stable particles and scattering channels relevant to the model are present.
2. Form-factor solution completeness: the axioms and operator data select the intended family of matrix elements, including normalization and polynomial ambiguities.
3. Spectral completeness: the included particle sectors and integration domain approximate the correlator to the claimed accuracy at the stated separation.

One level does not imply the next. An exact S matrix does not list every local operator, and exact low-particle form factors do not make a truncated correlator exact.

The integrability exact-data chain shows the independent spectral-tail check after the bootstrap. The exact and rigorous status comparison keeps exact matrix elements separate from unproved operator or series completeness.

Common pitfalls

Suppressing the locality phase. Cyclicity for a disorder or soliton field differs from that of a mutually local operator. The phase is part of the operator definition.

Importing a residue coefficient alone. State normalization, rapidity order, charge conjugation, and statistics determine the coefficient and sign. Translate the whole convention.

Calling a truncated spectral sum exact. Exact integrands can still leave an uncontrolled tail. Quote the separation range and a sector-by-sector stability test.

Exercises

1. Verify that $F_{2,\min}(\theta)=C\sinh(\theta/2)$ satisfies Watson exchange and cyclicity for $S=-1$ and $\ell_{\mathcal O}=1$.

Solution

Oddness gives $F_{2,\min}(\theta)=-F_{2,\min}(-\theta)$, exactly Watson’s equation. Moreover,

$$\sinh\!\left(\frac{\theta+2\pi i}{2}\right) =\sinh\!\left(\frac{\theta}{2}+i\pi\right) =-\sinh\!\left(\frac{\theta}{2}\right) =F_{2,\min}(-\theta)/C,$$

so cyclicity also holds.

2. Explain why $\Delta_{N+1}(r)$ is not generally an upper bound on the omitted tail $\sum_{n>N}C_n(r)$ even when every $C_n(r)$ is nonnegative.

Solution

Nonnegativity controls the sign but not the rate at which sectors decrease. The later terms could be individually smaller than $\Delta_{N+1}$ yet numerous enough to produce a larger sum, or could decrease nonmonotonically. An upper bound requires additional information, such as a proved ratio bound, while $\Delta_{N+1}$ alone is an empirical convergence diagnostic.

References

• Babujian, Hratchya M., Angela Foerster, and Michael Karowski. “The Form Factor Program: A Review and New Results—the Nested SU(N) Off-Shell Bethe Ansatz.” SIGMA 2 (2006): 082. DOI. Open PDF.
• Karowski, M., and P. Weisz. “Exact Form Factors in (1+1)-Dimensional Field Theoretic Models with Soliton Behaviour.” Nuclear Physics B 139 (1978): 455–476. DOI.