Lattice QCD is a formulation of quantum chromodynamics in which continuous spacetime is replaced by a discrete grid. It provides a nonperturbative framework for studying the strong interaction between quarks and gluons, especially at energies where expansions in a small coupling are unreliable. Numerical simulations calculate properties of hadrons directly from the underlying theory, with uncertainties assessed through statistical analysis and controlled limits of lattice spacing and volume. The grid is a computational regulator, not a claim that physical spacetime is fundamentally discrete. (pdg.lbl.gov)
Physical motivation and origins
QCD is the strong-interaction component of the Standard Model of particle physics. Through asymptotic freedom, its coupling becomes weaker at short distances, making perturbation theory effective for many high-energy processes. At hadronic distances, however, the coupling grows, and phenomena such as color confinement require different methods. Lattice calculations address this regime without expanding observables in powers of the strong coupling. (pdg.lbl.gov)
Kenneth Wilson’s 1974 paper “Confinement of quarks” established a lattice formulation of gauge theory preserving local gauge invariance. Wilson demonstrated confinement in a strong-coupling lattice expansion, connecting a gauge-invariant loop observable with the force between static color sources. This supplied an important theoretical argument and a framework for numerical investigation, rather than a general proof of confinement in continuum QCD. (journals.aps.org)
Fields and lattice formulation
Most large-scale simulations use a four-dimensional Euclidean lattice, where the time coordinate is continued to imaginary time. This connects the calculation of quantum expectation values with techniques from statistical mechanics. The lattice spacing (a) supplies a short-distance cutoff, while the number of sites determines the finite physical volume. Quark fields occupy sites; gauge variables representing gluonic parallel transport occupy links joining neighboring sites. (arxiv.org)
Each link variable is an element of the color gauge group SU(3), represented by a unitary (3\times3) matrix with determinant one. Products of link variables around closed paths yield gauge-invariant observables. The Wilson loop is one such observable; the simplest square loop, or plaquette, enters the Wilson gauge action. Local gauge invariance can therefore be retained exactly even at nonzero lattice spacing. (arxiv.org)
Quarks introduce additional complications because they are fermions. A naive discretization produces unwanted extra species, known as fermion doubling. Wilson, staggered, domain-wall, and overlap formulations handle this problem differently, with trade-offs involving computational expense, discretization effects, and the treatment of chiral symmetry. Agreement after taking the continuum limit is an important test of these distinct formulations. (arxiv.org)
Numerical calculation
The Euclidean path integral expresses observables as averages over field configurations. After quark fields are integrated out, the statistical weight includes the gauge action and quark determinants. These determinants incorporate dynamical, or “sea,” quarks. Omitting their effects defines the historically important quenched approximation, which is not full QCD. (arxiv.org)
For suitable parameters, ensembles of gauge configurations are generated using Markov chain Monte Carlo, often with hybrid Monte Carlo algorithms. Quark propagation on each configuration requires solving large sparse linear systems. These operations account for much of the computational workload and motivate extensive use of parallel computing and graphics processing units. GPU libraries such as QUDA provide specialized solvers and support calculations distributed across multiple devices. (arxiv.org)
Hadron energies are obtained from correlation functions of operators carrying the relevant quantum numbers. Their dependence on Euclidean time contains contributions from the ground state and excited states. Two-point functions determine spectra; three-point functions provide matrix elements, including those needed for decay amplitudes and form factors. Reliable extraction requires controlling excited-state contamination and statistical correlations. (pdg.lbl.gov)
Continuum limits and uncertainties
A simulation at one lattice spacing and volume is not automatically a continuum prediction. Calculations use several spacings to extrapolate toward (a=0), examine finite-volume effects, and tune quark masses. A measured reference quantity establishes the conversion from lattice units to physical units. Operators may also require renormalization and matching to specified continuum conventions. (arxiv.org)
Uncertainty budgets distinguish statistical fluctuations from systematic effects, including discretization errors, finite volume, quark-mass tuning, and correlation-function fitting. Configurations generated consecutively are not necessarily independent, so autocorrelations must be considered. Improved actions reduce leading discretization effects, but do not eliminate the need for continuum checks. (arxiv.org)
Applications and limitations
Major applications include hadron masses, quark masses, the strong coupling, and weak-decay matrix elements. Combining calculated decay constants and form factors with experimental measurements permits determinations of quark-mixing parameters. Lattice calculations also evaluate hadronic contributions to the anomalous magnetic moment of the muon, providing inputs to precision Standard Model tests. (pdg.lbl.gov)
At finite temperature, the Euclidean temporal extent controls the thermal ensemble: in natural units, (T=1/(aN_\tau)). Simulations determine the equation of state and properties of quark–gluon plasma. For physical quark masses and vanishing baryon chemical potential, calculations establish a smooth thermal crossover rather than a first-order transition. (arxiv.org)
At nonzero real baryon chemical potential, the quark determinant generally becomes complex. The resulting sign problem prevents its straightforward interpretation as a sampling probability. Taylor expansions around zero chemical potential and simulations at imaginary chemical potential offer restricted access to finite-density physics. Real-time dynamics also poses difficulties because ordinary simulations produce Euclidean correlation functions, from which real-time information must be reconstructed. (arxiv.org)