Approximation in Lattice Field Theories

Thomas Banks, Anish Suresh · 2025

Read Approximation in Lattice Field Theories by Thomas Banks, Anish Suresh (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-08501.txt.

Introduction

Thomas Banks and Anish Suresh check an approximation to a hard set of equations on a tiny lattice in a 2025 paper. The equations are the Schwinger-Dyson equations for an electron fluid. They govern how the fluid's parts correlate. In full, they form an endless chain. A large-N limit cuts the chain down to equations for one- and two-point values.

That cut needs a test. Monte Carlo runs on two-by-two and three-by-three lattices supply the test values. The derived equations match those values at large N, to the stated precision. Small, exact checks like these are how bold approximation schemes earn trust. Only then do they graduate to real systems that no one can solve head-on.

Catalog record

CREATORThomas Banks, Anish Suresh
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE43 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-08501

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Approximation in Lattice Field Theories Thomas Banks, Anish Suresh Abstract We investigate an approximation to the Schwinger-Dyson (SD) equations of the collective Coulomb field of the large N N homogeneous electron fluid. The large N N approximation transforms the infinite SD hierarchy is into a set of closed, equations for 1 and 2-pt correlators. In this paper, the dynamics of a toy model - a small square Euclidean lattice with periodic boundary conditions - are considered. The Markov Chain Monte Carlo numerical method evaluated the 1 and 2-pt correlation functions on a 2 × 2 2\times 2 and 3 × 3 3\times 3 lattice. The derived equations are checked with the correlator values, and an agreement at N ∼ 10 3 N\sim 10^{3} to order 10 − 3 10^{-3} was found. The agreement can be further strengthened by increasing runs in the Markov Chain Monte Carlo method. 1 Introduction In [ 1 ] , one of us presented a method for re-summing perturbative and vector 1/N expansions by using Schwinger-Dyson (SD) equations. The basic problem with SD equations is that they form an infinite hierarchy. The observation of [ 1 ] was that, when written in terms of one particle irreducible (1PI) correlation functions, in a model with a small coupling, this hierarchy could be systematically truncated by noting that the k point 1PI correlator vanishes like g^{k} . Each level of truncation gives rise to a finite closed system of non-linear integral equations, which are amenable to solution on modern computers. The same methods apply to the SD equations for the collective fields of vector 1/N expansions, where they are a generalization and improvement of a technique proposed in [ 2 ] . The purpose of the present note is to explore the accuracy of the proposed approximation on simple finite integrals, where we can perform both an “exact” evaluation of relevant correlators, and solve the approximate equations. We choose a model that is a cartoon of the homogeneous electron fluid, in which the electrons propagate on a 2\times 2 or 3\times 3 lattice. This model is treated in the re-summed 1/N expansion, where we choose the truncation of the SD equations in which only the single fermion loop contribution to the three and four point functions of the collective Coulomb field are kept. The unknown correlators that are solved for are the one and two point functions. The paper is structured as follows. In Section 2 , we derive the truncation of the infinite hierarchical equations up to order 1/N . Then, we describe the numerical method needed to evaluate the correlators in Section 3 . We also thoroughly explain the specifics of the numerical technique when applied to our problem. Finally, the results of the code, along with the code, are presented in Section 4 . We find agreement at the 10^{-3} accuracy level for N\sim 10^{3} on the 2\times 2 lattice and on the 3\times 3 lattice. While this accuracy is already strong, we also explain how to increase it. 2 Derivation of Closed Equations z_{11} z_{21} z_{11} z_{12} z_{22} z_{12} z_{11} z_{12} z_{11} Figure 1 : A 2\times 2 grid with periodic boundary conditions. The sites only connected via dashed lines represent mirrored values, so there are only 4 unique sites to consider. When generalized, this means that a \sqrt{M}\times\sqrt{M} grid will only possess M unique sites to consider. The value of the singlet boson field at each of the M independent lattice sites, \phi_{i} , is simply the stationary point \Phi_{0} plus a quantum fluctuation term. That is, \phi_{i}=\Phi_{0}+x(i)/\sqrt{N}. These sites are located on a \sqrt{M}\times\sqrt{M} Euclidean lattice with periodic boundary conditions (see Figure 1 ), making the system translational invariant. This also allows us to find a solution for \Phi_{0} that is independent of the lattice site. It is evident from the definition of \phi_{i} that the classical regime is recovered when N\rightarrow\infty . The action of this model is the following: S=N\bigg[\frac{1}{2}\sum_{i=1}^{M}\left(\Phi_{0}(i)+\frac{x(i)}{\sqrt{N}}\right)^{2}+\text{Tr}\left(\log{\left(A_{ij}+\frac{x(i)}{\sqrt{N}}\mathbb{1}_{ij}\right)}\right)\bigg]. (1) Throughout this manuscript, A_{ij}\equiv K_{ij}+\Phi_{0}(i)\mathbb{1}_{ij} , K_{ij} is the M\times M 2D discrete Laplacian matrix with periodic boundary conditions, and (A^{-1})_{ij} is the propagator between lattice sites i and j . We can then expand the action perturbatively, since we are interested in the terms that dominate in the action for large N . Ignoring the constant and linear terms, the action takes the form: S=\sum_{k=2}^{\infty}\frac{N^{1-k/2}}{k!}\sum_{i_{1},\cdots,i_{k}}^{M}S^{(k)}(i_{1},\cdots,i_{k})x(i_{1})\cdots x(i_{k}), (2) where S^{(2)}(i,j)=\delta_{ij}-(A^{-1})_{ij}(A^{-1})_{ji} and S^{(k)}(i_{1},\cdots,i_{k})=(-1)^{k}(k-1)!(A^{-1})_{i_{1}i_{2}}(A^{-1})_{i_{2}i_{3}}\cdots(A^{-1})_{i_{k}i_{1}} . The perturbative expansion provides a straightforward method to determine the stationary point, which we will cover in detail in Section 4 . Figure 2 contains the Feynman diagrams for S^{(2)},S^{(3)}, and S^{(4)} . Diagrams for higher order terms simply follow this pattern. S^{(2)} S^{(3)} S^{(4)} Figure 2 : Feynman diagrams for S^{(2)} , S^{(3)} , and S^{(4)} . The propagator here is A^{-1}_{ij} . We can then write the partition function in the imaginary time formalism/after Wick rotation: Z=\frac{1}{Z_{0}}\int[dx]\exp{\left(-S-\sum_{i}^{M}x(i)j(i)\right)}=e^{-W}. (3) The chain rule with any fixed i states that \partial_{x(i)}Z=0 . Only keeping terms up to order 1/N and rewriting the partition functions with respect to W and its derivatives, we arrive at the following M equations (one for each i ): \begin{split}j(i)=&\sum_{k}^{M}S^{(2)}(i,k)W_{k}+\sum_{k,l}^{M}\frac{1}{2\sqrt{N}}S^{(3)}(i,k,l)(W_{k}W_{l}-W_{kl})\\ &-\sum_{k,l,m}^{M}\frac{1}{6N}S^{(4)}(i,k,l,m)\bigg[W_{kl}W_{m}+W_{km}W_{l}+W_{lm}W_{k}-W_{klm}-W_{k}W_{l}W_{m}\bigg]\end{split} (4) where subscripts i_{1},\cdots,i_{k} on W indicate derivatives with respect to j(i_{1})\cdots j(i_{k}) . Taking the derivative with respect to j(t) on both sides produces another M^{2} equations (one for each (i,t) pair): \begin{split}0=&\sum_{k}^{M}S^{(2)}(i,k)W_{kt}+\frac{1}{2\sqrt{N}}S^{(3)}(i,k,l)\bigg[W_{kt}W_{l}+W_{k}W_{lt}-W_{klt}\bigg]-\delta_{it}\\ &-\frac{1}{6N}S^{(4)}(i,k,l,m)\bigg[(W_{klt}W_{m}+W_{kl}W_{mt}+\text{symmetric terms})-W_{klmt}-(W_{kt}W_{l}W_{m}+\text{symmetric terms})\bigg].\end{split} (5) Only M(M+1)/2 out of the M^{2} equations are unique, since derivatives commute. Eqns. ( 4 ) and ( 5 ), which consist of M+M(M+1)/2 equations in total, serve as the metrics for how successful this approximation is. Note that the sources j(i) are set to 0 in eqn. ( 4 ) by definition. 2.1 Approximating Higher Order Correlation Functions By definition, we have that W_{i} is the 1-pt correlation function \langle x(i)\rangle . There are only M such terms, so these terms are manageable. Similarly, calculating \langle x(i)x(j)\rangle gives us W_{ij} . However, working with 3 and 4-pt correlation functions is a lot more computationally intensive, especially given the large number of degrees of freedom. W k l m \Gamma W W W k l m b c a Figure 3 : A tree diagram showing the relationship between W_{3} and \Gamma_{3} . For large enough N , G_{3}=S^{(3)} . Using a Legendre transform \Gamma[s_{c}] , we can find an easy approximation for these functions. We may define the transform as: \Gamma[s_{c}]=-\sum_{i}s_{c}(i)j(i)+W[j]. (6) Introducing the Legendre transform is beneficial because the n th derivative of \Gamma[s_{c}] (with respect to the s_{c} ’s) is simply S^{(n)} (at leading order in N ). This fact, combined with the definition above, allows us to state the following: \begin{split}W_{3}(k,l,m)\equiv W_{klm}&=\frac{\partial^{3}W}{\partial j(k)\partial j(l)\partial j(m)}=-\frac{\partial}{\partial j(m)}(\Gamma_{kl}^{-1})=-\sum_{a}^{M}\frac{\partial\phi(a)}{\partial j(m)}\frac{\partial}{\partial\phi(a)}(\Gamma_{kl}^{-1})=\sum_{a,b,c}^{M}W_{am}W_{bk}\Gamma_{abc}W_{cl}\\ &\approx\sum_{a,b,c}^{M}W_{am}W_{bk}S^{(3)}(a,b,c)W_{cl}.\end{split} (7) Above, \Gamma^{-1}_{kl}=(\partial^{2}\Gamma/\partial s_{c}(k)\partial s_{c}(l))^{-1} - a fact that directly follows from eqn. ( 6 ). Similarly, W_{klmn} can be expressed as: \begin{split}W_{4}(k,l,m,n)\equiv W_{klmn}=&\sum_{a,b,c}^{M}\bigg[W_{amn}W_{bk}S^{(3)}(a,b,c)W_{cl}+W_{bkn}W_{am}S^{(3)}(a,b,c)W_{cl}+W_{cnl}W_{bk}S^{(3)}(a,b,c)W_{am}\bigg]\\ +&\sum_{a,b,c,d}^{M}W_{dn}W_{am}W_{bk}S^{(4)}(a,b,c,d)W_{cl}.\end{split} (8) A tree diagram representation of eqns. ( 7 ) and ( 8 ) can be found in Figures 3 and 4 . In the derivations up until now, one might have argued that one could ignore the 1/N term, since we are considering the large N limit. However, we included W_{4} to account for its potentially

Read the complete text