[Paper Review] Determinantal formulae and loop equations

M. C. Bergère, Bertrand Eynard|ArXiv.org|Jan 21, 2009

Random Matrices and Applications19 references37 citations

TL;DR

This paper establishes that correlation functions derived from the Christoffel-Darboux kernel of any second-order linear differential system satisfy loop equations, extending the well-known duality between determinantal formulae and loop equations beyond random matrix models. The key contribution is proving that the bosonic correlation functions $ W_n $, constructed via a determinantal formula from the kernel $ K(x_i,x_j) $, fulfill the loop equations universally, even in the absence of an underlying matrix model, via a generalization of the Sato-Heine exponential formula.

ABSTRACT

We prove that the correlations functions, generated by the determinantal process of the Christoffel-Darboux kernel of an arbitrary order 2 ODE, do satisfy loop equations.

Motivation & Objective

• To extend the duality between determinantal formulae and loop equations beyond random matrix models.
• To establish that correlation functions derived from the Christoffel-Darboux kernel of any second-order linear differential system satisfy loop equations.
• To generalize the Sato-Heine exponential formula for the kernel $ K(x,y) $ to arbitrary 2nd-order ODEs.
• To demonstrate that the loop equations are satisfied independently of the matrix model framework, relying only on the differential system structure.
• To provide a unified framework where determinantal structure and loop equations coexist in a broader class of systems.

Proposed method

• Define a $ 2 \times 2 $ linear differential system $ \Psi' = \mathcal{D} \Psi $ with traceless $ \mathcal{D}(x) $, leading to a spectral curve $ \hat{\mathcal{E}}(x,y) = \det(y - \mathcal{D}(x)) $.
• Construct the Christoffel-Darboux kernel $ K(x_1,x_2) = \frac{\psi(x_1)\tilde{\phi}(x_2) - \tilde{\psi}(x_1)\phi(x_2)}{x_1 - x_2} $ from the fundamental solution $ \Psi $.
• Define connected correlation functions $ W_n(x_1,\dots,x_n) $ via a cycle-sum formula involving $ K(x_i,x_{\sigma(i)}) $, with $ W_1 $ as a residue limit.
• Introduce non-connected correlators $ \mathcal{W}_n $ via a bracketed determinant $ \mathcal{W}_n = \left\langle \det K(x_i,x_j) \right\rangle $, relating to fermionic correlators.
• Prove that the $ W_n $ satisfy the loop equations using residue calculus and recursive differentiation, particularly showing $ \delta_y Q_2(x;x_1) = 0 $ implies the loop equation for $ n=0,1 $.
• Generalize the Sato-Heine formula $ K(x,y) = \exp\left( \sum_n \frac{1}{n!} \int_y^x \cdots \int_y^x W_n \right) $ to arbitrary 2nd-order systems.

Experimental results

Research questions

• RQ1 Do correlation functions derived from the Christoffel-Darboux kernel of an arbitrary second-order ODE satisfy loop equations, independent of a matrix model?
• RQ2 Can the Sato-Heine exponential formula for the kernel $ K(x,y) $ be extended to systems defined by arbitrary 2nd-order linear ODEs?
• RQ3 Is the duality between determinantal formulae and loop equations preserved when the underlying structure is a differential system rather than a random matrix?
• RQ4 How do the connected correlation functions $ W_n $, defined via cycle sums of the kernel, relate to the loop equations in the absence of a matrix measure?
• RQ5 What is the role of the spectral curve $ \hat{\mathcal{E}}(x,y) $ in ensuring the consistency of the loop equations for such systems?

Key findings

• The correlation functions $ W_n $, defined via cycle sums of the Christoffel-Darboux kernel $ K(x_i,x_j) $, satisfy the loop equations for all $ n \geq 0 $, even without a matrix model.
• For $ n=0 $, the loop equation reduces to $ W_2(x,x) + W_1(x)^2 = -\det \mathcal{D}(x) = \frac{1}{2} \operatorname{Tr} \mathcal{D}(x)^2 = -P_1(x) $, confirming consistency at the lowest order.
• For $ n=1 $, the loop equation is verified by applying $ \delta_y $ to the $ n=0 $ case, yielding $ W_3(x,x,y) + 2W_1(x)W_2(x,y) = -P_2(x;y) - \frac{\partial}{\partial y} \frac{W_1(x) - W_1(y)}{x - y} $.
• The proof relies on residue calculus and the identity $ \delta_y Q_2(x;x_1) = 0 $, which implies the loop equation holds for $ n=0 $ and $ n=1 $, and is extended to higher $ n $ via induction.
• The exponential formula $ K(x,y) = \exp\left( \sum_n \frac{1}{n!} \int_y^x \cdots \int_y^x W_n \right) $ holds universally for such systems, generalizing the Sato-Heine formula beyond matrix models.
• The loop equations are derived purely from the differential system structure, without assuming a matrix measure, proving their universality in this class of systems.