Questions: Lattice QCD Basics

Question 1

In lattice QCD, continuous spacetime is replaced by a discrete grid with lattice spacing a. Quark fields live on the lattice sites and gluon fields live on the links between sites (as SU(3) matrices U_mu(x)). Why are gluon fields placed on links rather than sites?

• A. Because there is not enough room at the sites for both quarks and gluons
• B. Because placing the gauge field on the link between sites x and x+a*mu-hat as a parallel transporter U_mu(x) = exp(i*g*a*A_mu(x)) preserves exact gauge invariance on the lattice — the Wilson plaquette action, built from products of link variables around elementary squares, is manifestly gauge-invariant and reduces to the continuum Yang-Mills action as a -> 0
• C. Because gluons are massless and need more space to propagate
• D. Because links have more degrees of freedom than sites

Answer: (B)

The link variable U_mu(x) transforms under a gauge transformation Omega(x) as U_mu(x) -> Omega(x) * U_mu(x) * Omega-dagger(x+mu). This means the product of links around a closed loop (a Wilson loop) is gauge-invariant, which is impossible to achieve with gluon fields placed on sites. The preservation of exact gauge symmetry at finite lattice spacing is Wilson's key insight (1974) and is crucial for the consistency of the lattice formulation. The plaquette action S = beta * sum_P (1 - 1/3 Re Tr U_P) reproduces the continuum action up to O(a^2) corrections.

Question 2

Lattice QCD calculations of hadron masses involve computing correlation functions C(t) = <0|O(t) O-dagger(0)|0> where O is an operator with the quantum numbers of the desired hadron. At large Euclidean time t, C(t) ~ exp(-m*t) where m is the hadron mass. Current lattice calculations reproduce the proton mass to within ~1%.

• T. True
• F. False

Answer: True

The correlation function at large t is dominated by the lightest state with the quantum numbers of O, and the exponential decay rate gives its mass. Modern lattice QCD calculations use dynamical quarks (including u, d, s, and often c quarks in the sea), physical pion masses, and multiple lattice spacings to extrapolate to the continuum limit. The BMW collaboration's 2008 calculation of the light hadron spectrum achieved ~1% agreement with experiment for the proton, neutron, and other light hadrons, demonstrating that QCD quantitatively explains the masses of the particles that make up most visible matter in the universe.

Question 3

Lattice QCD calculations require enormous computational resources. What are the main sources of computational cost, and what systematic uncertainties must be controlled?

Model answer: The main computational costs are: (1) Generating gauge field configurations by Monte Carlo sampling of the path integral, which requires inverting the Dirac operator (a large sparse matrix) for each dynamical quark flavor — the cost scales as (1/a)^4 * (L/a)^4 * (m_pi/m_pi,phys)^{-alpha} where a is the lattice spacing, L is the box size, and alpha depends on the algorithm. (2) Computing quark propagators on these configurations for the observables of interest. The main systematic uncertainties are: (a) discretization errors from finite a (controlled by using multiple a values and extrapolating to a = 0), (b) finite volume effects from the periodic box (controlled by using L >> 1/m_pi, typically L*m_pi > 4), (c) unphysical quark masses (older calculations used heavier-than-physical pion masses and extrapolated; modern calculations simulate at or near physical masses), and (d) renormalization of operators from lattice to continuum schemes.

The computational cost of lattice QCD has driven the development of specialized hardware (APE, QCDOC, Blue Gene, GPUs) and algorithmic advances (HMC, domain decomposition, multigrid solvers). Moore's law and algorithm improvements have enabled progress from quenched calculations (no dynamical quarks) in the 1980s to full dynamical 2+1+1 flavor calculations at physical pion masses today.