Lattice Regulators and Continuum Targets

A spacetime lattice becomes a formulation of quantum field theory only after the finite system, the target continuum theory, and the limiting claim have all been specified. This chapter develops that specification from the ground up, following the finite-system and continuum logic presented systematically by Montvay and Münster 1994, chs. 1–5. Its recurring example is a massive real scalar on a finite Euclidean lattice: simple enough to solve exactly, but rich enough to expose lattice spacing, volume, boundary, anisotropy, zero-mode, positivity, tuning, and symmetry-restoration questions separately.

The seven pages form a diagnostic sequence rather than a checklist to be completed mechanically. Start with the scientific question you need to answer, then choose the page that supplies the missing part of the regulator-to-continuum argument.

Choose the missing part of the formulation

If you need to decide…	Begin with…	Result you should be able to produce
Whether a finite lattice model actually specifies a target QFT	Lattice Regulators and Target Continuum Theories	A complete target–regulator–observable–limit statement
Which momenta, images, edge terms, or scale ratios follow from the geometry	Lattice Geometry, Boundaries, and Anisotropy	Physical extents, allowed modes, boundary terms, and an ordered limit plan
How to turn a continuum scalar action into a local lattice quadratic form	Scalar Lattice Actions and Difference Operators	Difference operators, summation by parts, the lattice field equation, and the leading cutoff term
Why the propagator contains trigonometric lattice momenta and how its poles move	Lattice Momentum, Propagators, and Cutoff Dispersion	A normalized finite-volume propagator and its continuum expansion
Which symmetries remain exact and how broken spacetime symmetries can be tested	Exact Symmetries, Broken Spacetime Symmetries, and Restoration	Regulator-symmetry channels, allowed mixings, tunings, and independent restoration tests
Whether Euclidean data admit a positive-state spectral interpretation	Reflection Positivity and Transfer-Matrix Criteria	A declared reflection, a positive-kernel test, and a precise statement of what has—and has not—been reconstructed
How simulations at several bare couplings are tied to one physical theory	Bare Parameters, Tuning Conditions, and Continuum Targets	Renormalized conditions defining a trajectory through bare-parameter space

The first four pages give the analytic core for a first lattice calculation. The final three prevent common logical shortcuts: exact finite-spacing symmetries do not imply restored continuum symmetry, a Euclidean action is not automatically associated with a positive Hamiltonian theory, and sending a symbol called $a$ to zero does not define a continuum trajectory.

The scalar benchmark that connects the pages

Take a $d$ -dimensional Euclidean hypercubic lattice with spacings $a_\mu$ , site coordinates $x_\mu=a_\mu n_\mu$ , and periodic extents $L_\mu=a_\mu N_\mu$ . For a real scalar, the nearest-neighbor action is

S_a[\phi] =\frac{a_0a_1\cdots a_{d-1}}{2} \sum_x\left[ \sum_{\mu=0}^{d-1}(\nabla_\mu^+\phi_x)^2 +m_0^2\phi_x^2 \right], \qquad \nabla_\mu^+\phi_x=\frac{\phi_{x+a_\mu\hat\mu}-\phi_x}{a_\mu}.

Periodic summation by parts gives the positive quadratic kernel

K_a(p)=m_0^2+\widehat p^2, \qquad \widehat p^2=\sum_\mu\frac{4}{a_\mu^2} \sin^2\!\left(\frac{a_\mu p_\mu}{2}\right),

at the discrete momenta $p_\mu=2\pi k_\mu/L_\mu$ . This one expression threads the chapter:

geometry and boundary conditions determine the mode set;
the difference operator determines $\widehat p^2$ ;
$K_a^{-1}$ is the finite-regulator two-point function;
the small- $a_\mu p_\mu$ expansion identifies cutoff effects;
pole locations in a distinguished Euclidean-time direction determine transfer energies;
comparing inequivalent lattice directions tests rotational restoration; and
holding renormalized dimensionless ratios fixed determines what $a\to0$ is meant to approach.

The benchmark is deliberately massive. At fixed $m>0$ , periodic-image effects are exponentially small for large $mL$ , whereas cutoff effects are powers of $am$ . The two error axes can therefore be varied independently. The massless zero mode is discussed as a failure test, not hidden by deleting one Fourier component.

Regulator data that must travel together

The chapter uses Euclidean signature, the weight $e^{-S_E}$ , and the site-wide conventions. A local lattice specification must add the data that the global convention cannot fix:

Surface	Minimum declaration	A check that can fail
Target	Dimension, fields, interactions, state or ensemble, and renormalized observable	Two regulators are compared using different observables
Geometry	Cell complex or lattice, spacings, extents, orientation, and boundary conditions	The quoted momentum set disagrees with the boundary phase
Finite variables	Site/link degrees of freedom, integration measure, bare parameters, and exact constraints	A Jacobian or boundary variable is omitted
Symmetry	Exact regulator group, broken target symmetries, anomalies, and allowed counterterms	A forbidden mixing is fitted or an allowed one is ignored
Scale hierarchy	Dimensionless combinations such as $am$ , $mL$ , and anisotropy $a_s/a_t$	A “continuum” sequence changes $mL$ unintentionally
Limits	What tends to zero or infinity, what is held fixed, and the observable defining convergence	Volume and critical limits are interchanged without a test
Validation	Analytic benchmark, independent formulation, Ward identity, positivity test, and error decomposition	Several plots share the same hidden approximation

This information is a scientific specification. It is not replaced by the name of a lattice action or by a list of bare couplings.

Preparation diagnostic

No single page is a hard prerequisite for the chapter overview. If the finite-dimensional integral itself is unfamiliar, review Regulated Bosonic Field Integrals. If the intended route is Hamiltonian rather than Euclidean, review Hamiltonian Initial Data and Phase Space. Probability limit theorems become essential only when ensemble estimates enter in later chapters; Probabilistic Convergence, Laws of Large Numbers, and Central Limit Theorems is the repair route.

You are ready for the analytic core if you can do three things:

vary a quadratic action while retaining boundary terms;
normalize a finite Fourier transform and invert its kernel; and
distinguish a dimensionful quantity from a dimensionless ratio that can be held fixed along a limiting sequence.

If only the third item is missing, begin with the first page rather than leaving the chapter.

Limits are claims about observables

The notation $a\to0$ is incomplete. A continuum claim is a statement that a renormalized observable $\mathcal O_R$ has a controlled limit along a tuned sequence,

\lim_{a\to0} \mathcal O_R\!\left(a,L(a),\mathbf g_0(a);\mu\right) =\mathcal O_R^{\mathrm{target}}(\mu),

where the trajectory $\mathbf g_0(a)$ is fixed by declared renormalized conditions and the volume prescription $L(a)$ is explicit. Different orders of limits can answer different questions. For a massive free scalar, taking $a\to0$ at fixed physical $L$ and then $L\to\infty$ agrees with the reverse order for suitable local correlators; near a critical point, with a zero mode, at finite temperature, or for nonlocal observables, that commutation must be established rather than assumed.

This distinction separates three levels of statement:

an identity at finite regulator, such as the exact inverse of a quadratic lattice kernel;
a controlled asymptotic statement, such as an $O(a^2)$ dispersion error at fixed physical momentum; and
a continuum-QFT claim, which additionally requires tuning, renormalized observables, restoration tests, and control of every relevant limit.

The general theory of regulator removal and universality belongs to Renormalization and EFT. This chapter supplies the lattice data needed to use that theory without hiding what was actually held fixed.

Review the chapter

The following five-part review tests construction and diagnosis together.

Check. For a periodic one-dimensional lattice, verify that $\sum_n f_n(\nabla^+g)_n=-\sum_n(\nabla^-f)_ng_n$ and identify the boundary term that appears on an open lattice.

Derive. Starting from the nearest-neighbor action, obtain $K_a(p)=m_0^2+4a^{-2}\sin^2(ap/2)$ and show that $K_a(p)=m_0^2+p^2-a^2p^4/12+O(a^4p^6)$ .

Compute. At $am=1/4$ and $ap=\pi/2$ , compare the lattice energy defined by $4\sinh^2(aE/2)=a^2m^2+4\sin^2(ap/2)$ with $\sqrt{m^2+p^2}$ . State which comparison is meaningful as a cutoff test.

Diagnose. A study holds $N=L/a$ fixed while reducing $a$ and calls the result an infinite-volume continuum limit. Explain which physical quantity also tends to zero and why the claim changes.

Synthesize. Write a one-page specification for a lattice theory of your choice containing target observables, regulator data, exact symmetries, broken symmetries, tuning conditions, scale hierarchy, limit order, and two independent validation routes.

Solutions and checkpoints

For the check, shift the periodic sum by one site. On an open chain the uncancelled endpoint contribution is $(f_{N-1}g_N-f_0g_0)/a$ in a convention with boundary values $g_0,g_N$ ; the precise form changes with which sites are dynamical, so that convention must be declared.

For the derivation, Fourier transforming $\phi_n=N^{-1/2}\sum_p e^{ipna}\widetilde\phi_p$ sends $\nabla^+$ to $(e^{iap}-1)/a$ . Its squared modulus is $4\sin^2(ap/2)/a^2$ . Expanding the sine gives the stated $-a^2p^4/12$ correction.

For the computation, insert dimensionless variables: $aE=2\operatorname{arsinh}\!\bigl(\tfrac12\sqrt{(1/4)^2+2}\bigr)$ , while $aE_{\rm cont}=\sqrt{(1/4)^2+(\pi/2)^2}$ . Their difference at this large lattice momentum is a legitimate finite-cutoff comparison but not an estimate of the leading small- $a$ coefficient; repeat at fixed physical $p$ with $ap\to0$ for that purpose.

For the diagnosis, $L=Na\to0$ . The sequence removes neither finite-volume effects nor the ultraviolet cutoff at fixed physical geometry; it approaches a shrinking box. An infinite-volume continuum sequence needs both $a\to0$ and $L\to\infty$ , with the target dimensionless ratios and their order specified.

The synthesis passes only if every limit names what is held fixed and every claimed continuum quantity is renormalized or demonstrably finite.

What this chapter prepares

After completing the chapter, you should be able to construct a regulator-to-continuum specification and to reject one whose geometry, observable, tuning, positivity, symmetry restoration, or limit order is missing. The next chapter, Lattice Observables and Continuum Inference, begins where this one stops: it turns correlation functions into renormalized observables, controls excited states and inverse problems, sets scales, improves actions and operators, and performs correlated continuum extrapolations.

References

Montvay, István, and Gernot Münster. Quantum Fields on a Lattice. Cambridge Monographs on Mathematical Physics. Cambridge University Press, 1994. doi:10.1017/CBO9780511470783.