Factorizable Correlation Functions

• Factorizable correlation functions are observables that decompose into products of local and nonlocal components, capturing inherent structural, topological, and algebraic features.
• They appear across diverse fields such as quantum many-body physics, quantum information, Toeplitz processes, conformal field theory, and statistics, providing key insights into critical scaling and finite-size effects.
• Their factorization enhances asymptotic analysis and simplifies the identification of universal behaviors, enabling efficient modeling in both statistical and quantum systems.

A factorizable correlation function is a correlation observable that, under certain structural or asymptotic limits, splits into a product of local and nonlocal components or allows representation in factorizable integral forms. This concept appears across quantum many-body physics, mathematical statistics, quantum information theory, and conformal field theory. The term captures precise factorization properties—often signaling essential structural, topological, or algebraic features—that distinguish these correlators from generic multi-point correlators.

1. Definitions and Structural Motivation

A factorizable correlation function is typically one where the correlation function $C_{r,N}$, depending on system size $N$ and chord distance $r$, can be written as

$C_{r,N} \sim R(\alpha) \, C_{r,\infty}$

in a specific limit ($N\to\infty$, $r\to\infty$, with $\alpha=r/N$ fixed), where $C_{r,\infty}$ is a "local" correlation function describing bulk critical scaling and $R(\alpha)$ is a "nonlocal" factor encoding finite-size, topological, or frustration effects (Li et al., 2018). This concept generalizes to correlation matrices and kernels in numerous contexts:

• Quantum channels and operator algebras: Factorizable quantum channels are those that admit a Stinespring dilation with an ancilla, enabling the channel to be written as an explicit partial trace over a larger Hilbert space (Musat et al., 2018).
• Toeplitz/determinantal processes: The correlation kernel is factorizable if it decomposes into a product/integral of functions each analytic on disjoint domains (Wiener–Hopf factorization) (Berggren et al., 2019).
• Quantum electron streams: In the time domain, single-particle states produce factorizable first-order correlation functions; multi-particle overlap destroys factorization (Moskalets, 2015).
• Statistical models: Explicit Cholesky factorization of correlation matrices in statistics yields a product of conditional structures, also qualified as "factorizable" (Madar, 2014).
• Conformal field theory: Correlators are factorizable with respect to the tensor category structure, constraints of ribbon Hopf algebras, and mapping class group invariance (Fuchs et al., 2013, Fuchs et al., 2012).

Factorizability thus encodes decomposition across locality, symmetry, topology, or algebraic structure.

2. Factorizable Correlators in Critical Spin Rings

In the transverse Ising ring and related models, factorizable correlation functions exhibit a precise structure in the thermodynamic and scaling limits. For a ring of $N$ spins, the conventional two-point function $$C_{r,N} = \langle E_0 | \sigma_j^\alpha \sigma_{j+r}^\alpha | E_0 \rangle$$ (with $\alpha=x,y,z$) can be factorized in the following scaling limit:

$$N\to\infty, \quad r\to\infty, \quad \alpha \equiv r/N \ \text{fixed}$$

yielding

$$C^{(\mathrm{O})}(\alpha) \equiv \lim_{\text{odd } N \rightarrow\infty} C_{r,N} = R(\alpha)\,C_{r,\infty}$$

where $C_{r,\infty} \sim r^{-\eta}$ is the bulk decay (with critical exponent $\eta$), and $R(\alpha)$ is a nonlocal factor determined by topology (e.g., ring frustration for odd $N$).

For the critical transverse Ising model ($\gamma=1,\,h=1$), explicit results show that the correlator factorizes into

$C_{r,2L+1} = (-1)^r S_{r,2L+1} B_1(\alpha)$

where $B_1(\alpha)=\cos(\pi\alpha/2)-\sin(\pi\alpha/2)$ acts as a universal "kink" or frustration factor characterizing topology-induced nonlocality (Li et al., 2018).

Similar structures arise for the isotropic XY and Heisenberg models, with model-dependent exponents and nonlocal factors $R^{(E)}(\alpha)$, $R^{(O)}(\alpha)$. The breaking of factorization in models with SU($n$), $n>2$, symmetry—e.g., SU(3) spin chains—is signaled by the necessity of higher-point correlators that cannot be collapsed into products of two-point objects (Ribeiro et al., 2018).

3. Factorizability in Operator Algebras and Quantum Channels

In operator algebraic quantum information theory, a quantum channel $T$ on $M_n(\mathbb{C})$ is said to be factorizable if it admits a Stinespring-type dilation over a finite von Neumann algebra $(N,\tau)$, i.e.,

$$T(x) = (\mathrm{id}_n \otimes \tau)\big(u(x\otimes 1_N)u^*\big), \qquad x \in M_n(\mathbb{C})$$

where $u$ is unitary in $M_n(\mathbb{C})\otimes N$ (Musat et al., 2018). Correlation matrices of projections and unitaries arising from traces $\tau(p_ip_j)$ and $\tau(u_i^*u_j)$ are called factorizable if they can be realized in this manner.

Intrinsically, factorization distinguishes channels and correlators that can be realized with finite-dimensional (type I) or infinite-dimensional (type II$_1$) ancillas. There exist explicit Schur multipliers (channels defined by entrywise product with a correlation matrix) with factorizable structure that can only be realized with infinite-dimensional ancilla, marking a uniquely infinite-dimensional phenomenon in quantum information theory.

Non-closure results for sets of such factorizable correlation matrices ($D_{\mathrm{fin}}(n)$, $F_{\mathrm{fin}}(n)$) highlight the subtlety: there are limiting correlation matrices (arising, for example, from projections with nonrational sums) which are factorizable as limits of finite-dimensional cases but cannot themselves be implemented with such ancillas (Musat et al., 2018).

4. Factorizable Kernels in Determinantal Point Processes

In determinantal processes defined via block Toeplitz minors, factorizable correlation functions emerge through contour-integral representations. The central step is the Wiener–Hopf factorization of the matrix-weight symbol in the Toeplitz matrix:

$\phi(z) = \phi_+(z)\phi_-(z)$

with $\phi_\pm$ analytic on disjoint annuli. In the infinite-size limit, the kernel takes a double contour-integral form:

$$K(x, y) = \frac{1}{(2\pi i)^2} \oint_{\Gamma_1} \oint_{\Gamma_2} \Phi(z, w)\;dz\,dw$$

where $\Phi(z, w)$ is built from the matrix factors and their inverses, each analytic in their domain (Berggren et al., 2019). The "factorized" nature of the kernel facilitates Riemann–Hilbert/saddle-point techniques for asymptotic analysis and clarifies universality and edge structure in random tiling models.

5. Factorizability in Correlation Functions for Logarithmic CFT

The notion of factorizability extends categorically in logarithmic conformal field theory (logCFT). Here, factorizability of the underlying ribbon category or Hopf algebra refers to the non-degeneracy of the canonical (coend-induced) Hopf pairing

$\omega_L: L\otimes L \to 1$

where $L=\int^{U\in\mathcal{C}} U^\vee \otimes U$ is the handle coend in the ribbon category $\mathcal{C}$ (Fuchs et al., 2013, Fuchs et al., 2012). This property ensures that bulk correlation functions—constructed via topological quantum field theory techniques—obey the requisite modular and mapping class group invariances.

Explicitly, the bulk partition function and higher-genus correlation functions can be written as bilinear combinations of characters (built from the Cartan matrix), or as universal morphisms in Hom-spaces over the modular category. Factorizability here is essential for consistency, modular invariance, and sewing (factorization) properties of the full CFT.

6. Factorizability and Cholesky Decomposition in Statistics

In mathematical statistics, explicit factorizable structures appear in the Cholesky decomposition of correlation matrices. A non-singular correlation matrix $R$ admits lower-triangular Cholesky factor $L$,

$$L_{ji} = \begin{cases} 0 & j < i \ \sqrt{1 - p_i R_{i-1}^{-1} p_i^T} & j = i \ \frac{\rho_{ij} - p_i R_{i-1}^{-1} p_j^T}{\sqrt{1 - p_i R_{i-1}^{-1} p_i^T}} & j > i \end{cases}$$

where $p_i$ is the vector of correlations and $R_{i-1}$ the principal submatrix (Madar, 2014). Alternatively, Cholesky factors can be written as square roots of determinant differences. In either case, the entries’ explicit factorizable form undergirds algorithmic matrix generation and statistical testing.

7. Conditions, Limitations, and Nonfactorizability

Factorizable correlation functions do not universally arise. In higher-rank integrable quantum chains (SU($n$) with $n>2$), the lack of crossing symmetry entails that higher-point correlators cannot be reduced to products of two-point functions—there are genuinely nonfactorizable correlators (Ribeiro et al., 2018).

In quantum electron optics, the first-order correlation function is factorizable in the time domain only when the wave-packets (levitons) are non-overlapping. Overlapping states give rise to nonfactorizable correlators, and energy-domain correlation functions can show accidental factorization that conceals the underlying multi-particle content (Moskalets, 2015).

A table summarizing domains and factorizability:

Domain	Factorizability Criteria	Key Feature
Critical spin rings	Thermodynamic + scaling limit	Nonlocal factors encode topology/frustration
Quantum channels (operator algebras)	Existence of ancilla/unitary dilation	Finite/infinite-dimensional realizability
Determinantal processes (Toeplitz)	Wiener–Hopf matrix symbol factorization	Double-contour integral kernels
Logarithmic CFT	Nondegenerate pairing in ribbon category	Consistent genus-$g$ correlators
Correlation matrices (statistics)	Explicit Cholesky factor via partials/determinants	Algorithmic generation, explicit form

• "Ring frustration and factorizable correlation functions of critical spin rings" (Li et al., 2018)
• "First-order correlation function of the stream of single-electron wave-packets" (Moskalets, 2015)
• "From non-semisimple Hopf algebras to correlation functions for logarithmic CFT" (Fuchs et al., 2013)
• "Non-closure of quantum correlation matrices and factorizable channels that require infinite dimensional ancilla" (Musat et al., 2018)
• "Correlation functions for determinantal processes defined by infinite block Toeplitz minors" (Berggren et al., 2019)
• "Higher genus mapping class group invariants from factorizable Hopf algebras" (Fuchs et al., 2012)
• "Correlation functions of the integrable SU(n) spin chain" (Ribeiro et al., 2018)
• "Direct formulation to Cholesky decomposition of a general nonsingular correlation matrix" (Madar, 2014)