Simulation of Lattice Gauge Theories

A lattice gauge simulation follows matter on sites and dynamical gauge fields on links, subject to a local relation between charge and electric flux. The main additional difficulty beyond an ordinary lattice model is preserving that relation while representing each link in a finite register. Exact gauge symmetry does not by itself imply an accurate cutoff approximation or a continuum field-theory prediction.

Required background. Lattice-model simulation provides encodings and local evolution; phase symmetry and gauge theory explains parallel transport and gauge redundancy. Helpful background. Lattice models introduces finite-size physics, and Hamiltonian simulation compares evolution algorithms.

Matter, links, and the physical subspace

The Hamiltonian formulation uses a spatial lattice and continuous time. Here $\hbar=1$; when discussing continuum length and mass scales, also set $c=1$. A finite model specifies the gauge group and representations, oriented lattice, boundary flux, matter discretization, link cutoff, and Hamiltonian coefficients. The initial state and measured observable complete the physical question. The following discussion concerns these regulated models; a continuum interpretation additionally needs parameter matching and convergence.

Let $G_x^a$ generate a local gauge transformation, with $a$ a Lie-algebra index. In the absence of external sources the physical states satisfy

$$G_x^a\lvert\psi_{\mathrm{phys}}\rangle=0, \qquad [H,G_x^a]=0$$

for every vertex and generator. Thus evolution preserves the physical subspace. For an Abelian group one can instead select fixed values $G_x\lvert\psi\rangle=g_x\lvert\psi\rangle$. Non-Abelian components do not generally commute, so arbitrary numbers $g_x^a$ cannot be assigned as simultaneous eigenvalues. Static non-Abelian charges carry representation spaces: include their charge operators in the total Gauss generators and impose the combined singlet condition.

A globally neutral state may still violate Gauss’s law at individual vertices. The two-site calculation below makes this distinction explicit.

Link transformations and the Hamiltonian

Link variables and transformations

For a compact group, a link $\ell=(x,i)$ points from $x$ to $x+\hat i$ and carries a parallel transporter $U_{x,i}$. Under local transformations $\Omega_x\in G$,

$$U_{x,i} \longmapsto \Omega_x U_{x,i}\Omega_{x+\hat i}^{\dagger}.$$

Matter in the corresponding representation transforms as

$$\psi_x \longmapsto \Omega_x\psi_x.$$

Therefore the nearest-neighbor hopping operator

$$\psi_x^\dagger U_{x,i}\psi_{x+\hat i}$$

is gauge invariant. Omitting $U_{x,i}$ would compare internal orientations at different sites without parallel transport.

Electric generators

Choose Hermitian matrices with $[T^a,T^b]=if^{abc}T^c$ and link generators obeying

$$[L^a,U]=T^aU,\qquad [R^a,U]=UT^a,$$ $$[L^a,L^b]=-if^{abc}L^c,\qquad [R^a,R^b]=if^{abc}R^c,\qquad [L^a,R^b]=0.$$

These relative signs follow from the Jacobi identity; left multiplication acts on the matrix indices in the opposite order when applied successively to the operator entries. The left and right Casimirs agree, $\sum_a(L^a)^2=\sum_a(R^a)^2$. Different links commute. This is the left/right convention used by Zohar, Cirac, and Reznik (2016); reversing a generator’s sign also changes its commutators and its contribution to Gauss’s law.

For compact $U(1)$, one may use an electric basis $\lvert\ell\rangle$, $\ell\in\mathbb Z$, with

$$E\lvert\ell\rangle = \ell\lvert\ell\rangle, \qquad U\lvert\ell\rangle = \lvert\ell+1\rangle,$$

so

$$[E,U]=U.$$

Plaquette holonomy

For an oriented elementary square in directions $i$ and $j$,

$$U_{p} = U_{x,i} U_{x+\hat i,j} U_{x+\hat j,i}^{\dagger} U_{x,j}^{\dagger}.$$

Its trace is gauge invariant because the vertex transformations cancel around the closed loop. In one spatial dimension there are no elementary spatial plaquettes, so a pure magnetic plaquette term is absent. This is one reason why 1+1-dimensional models are substantially simpler than higher-dimensional gauge theories.

Generic Kogut–Susskind structure

Suppressing convention-dependent lattice-spacing factors, a Hamiltonian with dynamical matter can be organized as

$$H = H_{\mathrm m} +H_{\mathrm{hop}} +H_E +H_B,$$

where

$$H_{\mathrm{hop}} = -w\sum_{x,i} \left( \psi_x^\dagger U_{x,i}\psi_{x+\hat i} +\mathrm{h.c.} \right),$$ $$H_E = \frac{g_E^2}{2} \sum_{\ell,a} (E_\ell^a)^2,$$

and a common magnetic term is

$$H_B = -\frac{1}{2g_B^2} \sum_p \left( \operatorname{Tr}U_p +\operatorname{Tr}U_p^\dagger \right).$$

The precise coefficients depend on spatial dimension, lattice spacing, group normalization, and whether constants have been subtracted. They must be copied from the declared regulated theory, not reconstructed from this schematic form.

Gauss’s law and local charge

Abelian form

For oriented $U(1)$ links and site charge $Q_x$, define

$$G_x = \sum_{\ell\,\mathrm{out\ of}\,x}E_\ell -\sum_{\ell\,\mathrm{into}\,x}E_\ell -Q_x.$$

The source-free physical subspace satisfies

$$G_x\lvert\psi_{\mathrm{phys}}\rangle=0.$$

Summing over all sites cancels internal link fluxes. For periodic boundaries this imposes total charge zero unless the formulation includes an appropriate background. With open boundaries, total charge equals net outward boundary flux. Global consistency is therefore a boundary statement as well as a local one.

Non-Abelian form

For a non-Abelian group, the generator at $x$ combines outgoing left fields, incoming right fields, and matter color charge,

$$G_x^a = \sum_i \left( L_{x,i}^a -R_{x-\hat i,i}^a \right) -Q_x^a,$$

where $Q_x^a=\psi_x^\dagger T^a\psi_x$. With the link convention above,

$$[G_x^a,G_y^b]=-i\delta_{xy}f^{abc}G_x^c.$$

Conjugation by $\exp(i\sum_{x,a}\alpha_x^aG_x^a)$ gives the displayed field and link transformations with $\Omega_x=\exp(i\alpha_x^aT^a)$. One may instead define $-G_x^a$, which has the positive Lie-algebra sign; the zero-Gauss physical subspace is unchanged. This overall choice must not be mixed with the opposite commutator convention halfway through a calculation. For $U(1)$, $L=R=E$ recovers the flux-divergence formula.

Constraint residuals

For an Abelian target sector $g_x$, a direct leakage diagnostic is

$$\epsilon_G = \frac1{N_v} \sum_x \left\langle (G_x-g_x)^2 \right\rangle.$$

It vanishes exactly only in the target sector. For non-Abelian groups, use a positive local Casimir residual, for example

$$\epsilon_G = \frac1{N_v} \sum_{x,a} \langle(G_x^a)^2\rangle$$

in the singlet sector. Report the normalization and cutoff dependence because the possible eigenvalues of $G_x^a$ depend on the link representation.

Let $K$ denote the sum of squared constraints and $P$ its zero-eigenvalue projector. If the smallest nonzero eigenvalue is $\Delta_G>0$, then

$$1-\operatorname{Tr}(P\rho) \le \frac{\langle K\rangle}{\Delta_G} =\frac{N_v\epsilon_G}{\Delta_G}.$$

Thus a small per-vertex residual need not imply small total leakage as the lattice grows. Even zero leakage is not sufficient for correct dynamics. Gauge-preserving coherent errors can alter physical dynamics while leaving every constraint exactly satisfied.

Explicit links or a constrained basis

Retain explicit gauge links

The most direct formulation assigns registers to matter sites and gauge links. Local matter hopping changes both occupation and adjacent flux, while electric and plaquette terms act on links. Advantages include manifest spatial locality, direct access to flux observables, and a transparent approach to higher dimensions.

The costs are substantial:

• continuous groups require finite link registers;
• Gauss’s law reduces the physical space but does not automatically reduce the allocated register;
• a plaquette gate acts on at least four links before matter or ancillas;
• non-Abelian links carry representation and magnetic indices;
• hardware noise can leave the physical subspace.

Solve Gauss’s law

In one spatial dimension with open boundaries, Abelian Gauss’s law can be solved recursively. If $E_0$ is the left boundary flux and sites are ordered,

$$E_n = E_0+\sum_{j=1}^{n}Q_j.$$

Substituting into the electric energy eliminates link variables,

$$H_E = \frac{g_E^2}{2} \sum_n \left( E_0+\sum_{j\le n}Q_j \right)^2.$$

The register becomes smaller and every represented state can satisfy the solved constraints. The price is a long-range Coulomb interaction among charges. Boundary flux and total-charge compatibility must still be enforced, and the strategy does not generalize so simply to loops, higher dimensions, or non-Abelian groups.

This is an important locality tradeoff: eliminating redundant gauge registers can make the physical Hamiltonian less local on the remaining degrees of freedom.

Use a gauge-invariant basis

Loop, string, hadron, spin-network, prepotential, or related bases solve local singlet constraints at the level of basis states. They can reduce unphysical storage and make gauge-invariant ansätze natural. However, the resulting basis may have nontrivial global constraints, irregular state counting, complicated operators, or less transparent spatial locality.

There is no universally best basis. Compare formulations at fixed physical volume, link cutoff, target observables, and precision. A smaller Hilbert space can require a more expensive Hamiltonian oracle or more complicated state preparation.

Finite link spaces and their algebras

Hard electric-field cutoff

For compact $U(1)$, retain

$$-\Lambda\le\ell\le\Lambda, \qquad d_{\mathrm{link}}=2\Lambda+1.$$

A binary register needs

$$n_{\mathrm{link}} = \left\lceil \log_2(2\Lambda+1) \right\rceil$$

qubits, plus a convention for unused computational states. Projecting the shift operator onto this interval gives

$$U_\Lambda\lvert\ell\rangle = \begin{cases} \lvert\ell+1\rangle,&\ell<\Lambda,\\ 0,&\ell=\Lambda. \end{cases}$$

The electric commutator remains exact:

$$[E_\Lambda,U_\Lambda]=U_\Lambda.$$

What fails is unitarity:

$$U_\Lambda^\dagger U_\Lambda=I-\lvert\Lambda\rangle\langle\Lambda\rvert,\qquad U_\Lambda U_\Lambda^\dagger=I-\lvert-\Lambda\rangle\langle-\Lambda\rvert.$$

Consequently $[U_\Lambda,U_\Lambda^\dagger]$ is the difference of the upper- and lower-edge projectors. The exact electric commutator preserves the continuous $U(1)$ transformation of the truncated link, so appropriately constructed hopping and plaquette terms can still satisfy Gauss’s law exactly. Their dynamics differs from the untruncated rotor. Increasing $\Lambda$ and checking both cutoff-edge weight and the desired observable tests that difference.

Cyclic finite group

On the same labels $-\Lambda,\ldots,\Lambda$, let $d=2\Lambda+1$ and close the raising shift with $U_d\lvert\Lambda\rangle=\lvert-\Lambda\rangle$. This operator is unitary, but

$$[E_\Lambda,U_d] =U_d-d\lvert-\Lambda\rangle\langle\Lambda\rvert.$$

Instead of the rotor commutator, it obeys the finite Weyl relation $ZU_d=e^{2\pi i/d}U_dZ$, where $Z=e^{2\pi iE_\Lambda/d}$. It can therefore define a $\mathbb Z_d$ gauge theory with modular Gauss constraints. It is not the same finite model as the hard cutoff. A relation to $U(1)$ requires a suitable sequence of Hamiltonians and observable convergence, not merely more levels.

The same three electric labels support two different link operators. The hard shift stops at the upper edge and preserves $[E,U]=U$. The cyclic shift wraps to the lower edge and is unitary, but obeys a finite Weyl algebra. These properties determine different gauge constraints and different finite-model dynamics.

Quantum-link and representation truncations

Quantum-link models replace infinite-dimensional link rotors by finite spin or rishon representations. For an Abelian example,

$$E=S^z, \qquad U\propto S^+.$$

Here $[E,U]=U$ still holds, whereas $U$ is not unitary and its matrix elements have spin-dependent weights. Gauss’s law can remain exact despite this change from rotor dynamics. For non-Abelian compact groups, a Peter–Weyl link basis can be truncated to a finite set of irreducible representations,

$$\mathcal H_{\mathrm{link}} \simeq \bigoplus_{r\in\mathcal R_\Lambda} V_r\otimes V_r^*.$$

The retained representation set, multiplication rules after projection, and closure of compiled operators must be stated. “Three qubits per link” does not identify the gauge-field approximation.

Cutoff convergence is observable-dependent

Useful diagnostics include

$$p_{\mathrm{edge}} = \sum_{\ell\in\partial\mathcal E_\Lambda} \langle\ell\rvert\rho_{\mathrm{link}}\lvert\ell\rangle,$$

changes in the final observable under $\Lambda\to\Lambda+1$, and stability of low-energy spectra or transition amplitudes. Strong electric coupling can suppress high flux, while weak coupling and continuum approaches can require larger link spaces. A cutoff adequate for ground energy may fail for a violent real-time process that generates broad flux distributions.

Preserving the physical subspace during evolution

Exact preservation by encoding

The strongest option is to represent only physical states. If an isometry $V:\mathcal H_{\mathrm{phys}}\to\mathcal H_{\mathrm{reg}}$ embeds the physical space, every logical operation should satisfy

$$U_{\mathrm{reg}}V = VU_{\mathrm{phys}}.$$

Gauge leakage is then excluded at the logical level. The tradeoff is that encoded operators can become nonlocal or expensive, and an invalid hardware operation can leave the code space unless the encoding detects it.

Exact preservation by primitive gates

With explicit links, construct every compiled generator $K_a$ so that

$$[K_a,G_x^b]=0 \qquad \text{for all }a,x,b.$$

Then each factor $e^{-iK_a\delta t}$ preserves the physical subspace at every intermediate time. A product formula assembled from these factors remains exactly gauge invariant even at finite step size,

$$[U_{\mathrm{PF}}(t),G_x^b]=0.$$

This requires compiling a gauge-invariant matter–link hop as one logical primitive or as a sequence whose full action and intermediate leakage are controlled. A product of exponentials of bare matter hopping and a separate link generator approximates the sum of those generators, not their coupled matter–link product. Such a split is not a valid product formula for the desired interaction. A correctly synthesized logical gate may itself use hardware gates that leave the physical subspace temporarily; only the complete ideal block is then guaranteed to preserve Gauss’s law.

Energy penalties

For an Abelian target sector use

$$H_{\mathrm{pen}}=\lambda\sum_x(G_x-g_x)^2.$$

For a non-Abelian singlet sector use $H_{\mathrm{pen}}=\lambda\sum_{x,a}(G_x^a)^2$, including any external-source operators in $G_x^a$. Arbitrarily shifting noncommuting components by classical numbers does not define a general color-charge sector.

For sufficiently large $\lambda$, gauge-violating states are energetically separated. This can suppress weak perturbative transitions, but it is not free error correction. A finite penalty modifies off-shell dynamics, increases the Hamiltonian norm and digital simulation cost, and can create calibration or spectral-resolution demands. Resonant noise can still cross the penalty gap.

Penalty convergence must be demonstrated in the final observable while checking both leakage and in-sector distortion. Taking $\lambda$ as large as hardware allows is not automatically optimal.

Linear protection and constrained dynamics

A protection term of the form

$$H_{\mathrm{prot}} = V\sum_x c_x(G_x-g_x)$$

can energetically distinguish gauge sectors with fewer-body controls in some Abelian implementations. The coefficients $c_x$ must separate the relevant violations; uniform coefficients may protect only total charge. Quantum-Zeno, dissipative, and dynamical-decoupling strategies can also confine evolution to a target sector under specific time-scale assumptions.

Every protection scheme needs a residual effective Hamiltonian. If $H=H_0+H_{\mathrm{err}}+H_{\mathrm{prot}}$, state the leading in-sector terms generated by virtual excursions, not only the observed suppression of $\epsilon_G$.

Detection and postselection

If all $G_x$ or commuting stabilizer-like functions of them are measurable, one can reject records outside the target sector. With acceptance probability $p_{\mathrm{acc}}$, obtaining $N_{\mathrm{acc}}$ accepted shots costs on average

$$N_{\mathrm{raw}} = \frac{N_{\mathrm{acc}}}{p_{\mathrm{acc}}}.$$

Postselection does not detect errors that act entirely within the physical space. It can also bias time traces if acceptance depends strongly on the state or observable. Report raw and conditioned results, acceptance with uncertainty, and the prespecified acceptance rule.

Compile the Gauge Hamiltonian

Mass and electric energy

Mass, staggered-density, and Abelian electric-energy terms are diagonal in occupation and electric bases. Binary encodings implement them through phase polynomials or arithmetic; unary and qudit encodings can use level-dependent phases. The resource depends on coefficient precision and link dimension, not only the number of terms.

Matter–gauge hopping

A gauge-covariant hop has the schematic action

$$\lvert n_x,e,n_y\rangle \longleftrightarrow \lvert n_x-1,e+\delta e,n_y+1\rangle,$$

where $\delta e$ is fixed by orientation, matter charge, and representation. The update must be coherent and conditional on valid occupations and link boundaries. In a binary encoding it generally requires arithmetic or a multi-register select operation. In a native qudit implementation it can be a mixed-dimensional controlled rotation.

The hopping sign also contains fermionic parity information. In more than one dimension, the Jordan–Wigner or alternative fermion map can add strings even when the gauge link is geometrically adjacent. Gauge locality and fermionic encoding locality both affect the compiled interaction.

Plaquette dynamics

The plaquette operator changes a closed loop of link flux. For $U(1)$,

$$U_p+U_p^\dagger$$

raises the oriented circulation around $p$ or lowers it. A digital implementation can use an ancilla to compute allowed shifts, apply a phase or rotation, and uncompute; direct qudit gates or analog ring exchange are other possibilities. Adjacent plaquettes share links, so plaquette scheduling is a hypergraph-coloring problem rather than ordinary edge coloring.

Product-formula groups

A useful split keeps each group gauge invariant,

$$H = H_{\mathrm{diag}} +H_{\mathrm{hop}}^{(1)} +\cdots +H_{\mathrm{hop}}^{(c_h)} +H_{\mathrm{plaq}}^{(1)} +\cdots.$$

Here direction, parity, or graph coloring creates internally parallel groups. The formula error is controlled by commutators among overlapping groups, while the exact Gauss symmetry survives finite step size if every group commutes with the generators. Trotter–Suzuki Methods develops the general error analysis.

Block encodings and fault-tolerant algorithms

Sparse-Hamiltonian, linear-combination, qubitization, and interaction-picture algorithms can improve asymptotic precision dependence. Their application cost is determined by the actual SELECT and PREPARE oracles for group-valued link updates, coefficients, plaquette products, and fermion signs. An oracle query is not one elementary gate.

The cutoff affects both the Hamiltonian norm and oracle arithmetic. A resource comparison must use the same physical volume, bare parameters, link digitization, target observable, and total error. Otherwise a favorable query count can describe a different regulated theory.

Prepare Physical States

Strong-coupling vacuum

When electric energy dominates, a simple vacuum often minimizes link flux and fills staggered matter sites according to the mass convention. It can be a product state satisfying Gauss’s law. This gives a valuable preparation starting point and a solvable validation limit.

The “vacuum” depends on regulator, boundary flux, mass sign, and background charge. A bare product vacuum is not the interacting continuum vacuum.

Charges and flux strings

A net Abelian charge is incompatible with zero-background Gauss’s law on a periodic lattice. A charge–anticharge pair can be prepared with a connecting Wilson line,

$$\mathcal M_{xy} = \psi_x^\dagger \left( \prod_{\ell\in\gamma_{xy}}U_\ell \right) \psi_y,$$

where the ordered path $\gamma_{xy}$ dresses the matter operators. Acting with $\mathcal M_{xy}$ changes endpoint charges and the intervening flux together, preserving Gauss’s law. This is the natural initial state for string-breaking experiments.

Adiabatic and variational preparation

One can interpolate from a tractable physical Hamiltonian,

$$H(s) =(1-s)H_{\mathrm{easy}}+sH_{\mathrm{target}},$$

provided every $H(s)$ preserves the desired gauge sector. The required time is set by finite-size gaps and matrix elements along the path. Near a phase transition or continuum critical point, the bottleneck can worsen with volume.

Gauge-invariant variational ansätze are built from exponentials of physical operators. Energy variance, Gauss residuals, exact small-volume comparisons, and held-out observables provide complementary checks. Low energy plus small gauge violation still does not establish the correct ground state if the physical spectrum is dense.

Scattering and wave packets

Real-time high-energy applications ultimately require interacting vacua, localized or momentum-resolved incoming packets, sufficient separation from boundaries, and an outgoing measurement protocol. State-preparation and finite-volume costs can dominate the central collision. A quench from an easy bare state probes interesting dynamics but is not automatically an $S$-matrix calculation.

Gauge-Invariant Observables

Local charge and electric flux

In an Abelian theory, $Q_x$ and $E_\ell$ are gauge invariant, and flux profiles reveal strings, screening, and boundary flow. In a non-Abelian theory, individual color components $Q_x^a$ or $E_\ell^a$ are covariant, not gauge invariant; neither is the square of one arbitrarily selected component in general. Physical local scalars include singlet matter density and the summed electric Casimir $\sum_a(E_\ell^a)^2$. An Abelian discrete-divergence check is

$$\left\langle \nabla_{\mathrm{lat}}\cdot E_x-Q_x \right\rangle = \langle G_x\rangle.$$

The stronger squared residual detects mixtures of positive and negative violations that cancel in the mean.

Wilson lines and loops

An open Wilson line is gauge covariant at its endpoints and becomes an observable when dressed with matter or external sources. A closed Wilson loop is

$$W(C) = \operatorname{Tr} \mathcal P \prod_{\ell\in C}U_\ell,$$

where $\mathcal P$ orders noncommuting link matrices along $C$. Loop expectations can diagnose flux and confinement properties, but finite loops on one lattice do not determine an asymptotic area law.

String breaking

Prepare separated external or dynamical charges connected by flux and monitor:

• electric field along and outside the initial path;
• charge density near the endpoints and newly created pairs;
• energy stored in field, matter, and interaction terms;
• connected correlations and entanglement;
• dependence on separation, mass, volume, cutoff, and boundaries.

A falling average flux alone can also result from gauge-violating leakage, dephasing, or boundary escape. Gauss residuals and energy accounting must accompany the physical interpretation.

Pair production and vacuum persistence

For an initial vacuum $\lvert\Omega_0\rangle$, one may measure a particle density and the Loschmidt or vacuum-persistence probability

$$P_{\mathrm{vac}}(t) = \left| \langle\Omega_0\rvert e^{-iHt} \lvert\Omega_0\rangle \right|^2.$$

The particle-number operator requires a regulator-specific vacuum subtraction or staggered convention. Pair production on a few sites is a finite regulated transition, not by itself a measurement of the infinite-volume Schwinger rate.

Spectra and correlation functions

Gauge-invariant interpolating operators create neutral hadronic, mesonic, glueball, or flux-loop excitations. Real-time correlators,

$$C_O(t) = \langle\Omega\rvert O^\dagger(t)O(0) \lvert\Omega\rangle,$$

can yield finite-volume energies after controlled reconstruction. A bare charged field correlator is gauge dependent unless gauge fixed or connected by a Wilson line. Measurement protocols should target physical operators from the start rather than reconstruct gauge-variant amplitudes and interpret them later.

Worked Example: Two-Site Schwinger Dynamics

The smallest nontrivial open staggered-fermion system makes the constraint and the link-elimination tradeoff exact. Work in units with $\hbar=1$. Two matter qubits occupy sites $1$ and $2$, one $U(1)$ link points from $1$ to $2$, and

$$Z\lvert0\rangle=\lvert0\rangle, \qquad Z\lvert1\rangle=-\lvert1\rangle.$$

Take $\lvert1\rangle$ to mean occupied, so $n_j=(I-Z_j)/2$. Choose negatively charged fermions and staggered background charges $(b_1,b_2)=(0,1)$, giving $Q_j=b_j-n_j$. Explicitly,

$$Q_1 = \frac{Z_1-1}{2}, \qquad Q_2 = \frac{Z_2+1}{2}.$$

With zero external boundary flux, Gauss’s law is

$$G_1=E-Q_1=0, \qquad G_2=-E-Q_2=0.$$

The neutral physical sector contains

$$\begin{aligned} \lvert\mathrm v\rangle &= \lvert01\rangle_{\mathrm m} \lvert E=0\rangle, \\ \lvert\mathrm p\rangle &= \lvert10\rangle_{\mathrm m} \lvert E=-1\rangle. \end{aligned}$$

The first is the bare staggered vacuum. The second contains a charge–anticharge pair and one unit of oriented electric flux.

Gauge-invariant Hamiltonian

Let

$$H_m = \frac m2(-Z_1+Z_2), \qquad H_E=\frac\kappa2E^2,$$

and

$$H_{\mathrm{hop}} = -w \left( \sigma_1^-U^\dagger\sigma_2^+ + \sigma_1^+U\sigma_2^- \right).$$

Here $\sigma^\pm=(X\pm iY)/2$; in this occupation convention $\sigma^-\lvert0\rangle=\lvert1\rangle$ creates a fermion. The first term moves the negatively charged fermion from site $2$ to site $1$: $Q_1$ decreases by one, $Q_2$ increases by one, and $U^\dagger$ lowers $E$ by one. Both Gauss equations remain satisfied.

In the ordered basis $\{\lvert\mathrm v\rangle,\lvert\mathrm p\rangle\}$,

$$H_{\mathrm{phys}} = \begin{pmatrix} -m&-w\\ -w&m+\kappa/2 \end{pmatrix}.$$

Introduce a logical Pauli operator with $Z_{\mathrm L}\lvert\mathrm v\rangle=\lvert\mathrm v\rangle$. Then

$$H_{\mathrm{phys}} = \frac\kappa4I -wX_{\mathrm L} -\left(m+\frac\kappa4\right)Z_{\mathrm L}.$$

Solving Gauss’s law has reduced three registers to one logical qubit. The electric energy has become a charge-dependent diagonal term.

Exact pair-production probability

Define

$$\Omega = \sqrt{ w^2+\left(m+\frac\kappa4\right)^2 }.$$

Starting from $\lvert\mathrm v\rangle$, the probability of finding the pair state is

$$P_{\mathrm p}(t) = \frac{w^2}{\Omega^2} \sin^2(\Omega t).$$

The electric energy and squared flux follow directly,

$$\langle E^2(t)\rangle = P_{\mathrm p}(t), \qquad \langle H_E(t)\rangle = \frac\kappa2P_{\mathrm p}(t).$$

At $w=0$, no pair is produced; the $\Omega=0$ expression is understood by continuity as zero. At resonance $m=-\kappa/4$ with $w\ne0$, the oscillation amplitude reaches one. Away from resonance, mass and electric energy suppress the transition. This is a finite two-level oscillation, not the infinite-volume pair-production rate.

What a bare matter hop gets wrong

If a compiler applies $\sigma_1^-\sigma_2^+$ without shifting the link, it produces

$$\lvert10\rangle_{\mathrm m}\lvert E=0\rangle.$$

For this state,

$$G_1=1, \qquad G_2=-1.$$

Total charge remains zero, yet Gauss’s law is violated at both sites. This single step demonstrates why global-charge monitoring cannot replace local constraint checks.

This small calculation checks charge and link signs, energy factors, and explicit-link versus reduced encodings. It cannot test spatial plaquettes, non-Abelian representations, or continuum scaling.

From a finite lattice to continuum physics

Distinguish the limits

For a box with $N$ sites along each side, write its physical length as $L_{\mathrm{phys}}=Na$. A regulated observable depends separately on

$$O(a,L_{\mathrm{phys}},\Lambda;\boldsymbol g_0),$$

where $a$ is lattice spacing, $\Lambda$ is the link cutoff, and $\boldsymbol g_0$ denotes bare parameters. A continuum limit tunes these parameters to hold selected renormalized quantities fixed as $a\to0$. An infinite-volume limit additionally takes $L_{\mathrm{phys}}\to\infty$, and the link approximation must converge in that regime. The order or joint scaling of these limits depends on the theory and observable.

Increasing the number of sites at fixed $a$ approaches infinite volume, not the continuum. Decreasing $a$ at fixed site count shrinks the physical box. Increasing link dimension at one bare coupling removes one digitization error but does not perform renormalization.

Scale setting and parameter matching

A dimensionful prediction needs a measured or computed quantity to set the lattice scale. Bare masses and couplings generally differ from renormalized ones. Useful dimensionless comparisons include

$$m_{\mathrm{gap}}L_{\mathrm{phys}}, \qquad a m_{\mathrm{gap}}, \qquad \frac{m_{\mathrm{hadron}}}{m_{\mathrm{ref}}}.$$

Track these ratios across lattice spacings and volumes. The regulator’s symmetry violations and operator renormalization must be included where relevant.

Real time does not remove regulator obligations

Quantum hardware naturally evolves in real time, avoiding the direct need to sample a complex real-time path-integral weight. That is a genuine motivation, but it does not remove:

• state preparation of the interacting vacuum or incoming particles;
• gauge-link and matter digitization;
• ultraviolet renormalization and scale setting;
• finite-volume recurrences and boundary reflections;
• long coherent evolution for spectral resolution;
• measurement and reconstruction of physical observables;
• error correction or validated mitigation.

“No sign problem on the quantum device” is not an end-to-end complexity proof.

Observable errors and computational cost

The finite-model approximation, Gauss-sector leakage, state preparation, evolution, hardware noise, and statistical inference affect different quantities. To combine errors rigorously, compare the same normalized observable at successive stages. If those expectations are $o_0,\ldots,o_k$, then

$$|o_k-o_0|\le\sum_{j=1}^k|o_j-o_{j-1}|.$$

Each increment needs its own bound or convergence study; a Gauss residual, gate infidelity, energy error, and sampling standard error cannot simply be added. In particular, replacing $H$ by a wrong but gauge-invariant $H+\Delta H$ leaves every Gauss check perfect. Compare small-system spectra and time traces, solvable limits, and observables not used to fit the model.

Explicit-link storage scales as $N_s n_{\mathrm m}+N_\ell n_{\mathrm g}$ qubits before ancillas. For a hard $U(1)$ cutoff, $n_{\mathrm g}=\lceil\log_2(2\Lambda+1)\rceil$. This does not include state preparation, coupled hopping, plaquette synthesis, fermion strings, or measurement repetitions. The state-dependent cutoff bounds of Tong and colleagues (2022), for example, assume controlled initial local quantum numbers and specified Hamiltonian growth; they are not guarantees for arbitrary high-flux states.

Classical tensor networks, exact diagonalization, and Euclidean lattice methods in suitable regimes provide complementary comparisons. Their limitations motivate quantum simulation but do not by themselves establish an end-to-end quantum advantage. Verification of quantum simulation develops these comparisons. For dated experimental progress and open problems, see lattice-gauge simulation research.

Exercises

1. Check Abelian gauge invariance of a hop

Let

$$\psi_x\mapsto e^{iq\alpha_x}\psi_x, \qquad U_{xy}\mapsto e^{iq\alpha_x}U_{xy}e^{-iq\alpha_y}.$$

Show that $\psi_x^\dagger U_{xy}\psi_y$ is gauge invariant, while $\psi_x^\dagger\psi_y$ is not for independent $\alpha_x$ and $\alpha_y$.

Solution

The dressed hop transforms as

$$\begin{aligned} \psi_x^\dagger U_{xy}\psi_y \mapsto{}& \psi_x^\dagger e^{-iq\alpha_x} e^{iq\alpha_x}U_{xy}e^{-iq\alpha_y} e^{iq\alpha_y}\psi_y \\ ={}& \psi_x^\dagger U_{xy}\psi_y. \end{aligned}$$

Without the link,

$$\psi_x^\dagger\psi_y \mapsto e^{iq(\alpha_y-\alpha_x)} \psi_x^\dagger\psi_y,$$

which is invariant only for a global transformation or special equal phases. The link is the parallel transporter between independently chosen local frames.

2. Derive global charge–flux consistency

Sum the Abelian Gauss law over every site of a finite oriented graph. Show what happens to internal links and state the result for periodic and open boundaries.

Solution

Every internal link contributes $+E_\ell$ at its tail and $-E_\ell$ at its head, so all internal contributions cancel. Thus

$$\sum_xG_x = \Phi_{\mathrm{out}} -\sum_xQ_x,$$

where $\Phi_{\mathrm{out}}$ is net outward boundary flux. In a physical sector with $G_x=0$,

$$Q_{\mathrm{tot}}=\Phi_{\mathrm{out}}.$$

A periodic lattice has no boundary contribution, hence total charge must vanish unless background charges or modified global sectors are included. On an open lattice, nonzero total charge is supported by boundary flux.

3. Solve the one-dimensional Gauss law

For an open Abelian chain with

$$G_n=E_n-E_{n-1}-Q_n=0,$$

derive $E_n$ in terms of the left boundary field and charges. Expand $\sum_{n=1}^{N-1}E_n^2$ and explain why link elimination creates long-range interactions.

Solution

Recursion gives

$$E_n = E_0+\sum_{j=1}^{n}Q_j.$$

Therefore

$$\sum_{n=1}^{N-1}E_n^2 = \sum_{n=1}^{N-1} \left( E_0+\sum_{j\le n}Q_j \right)^2.$$

On expansion, a pair $Q_jQ_k$ appears in every term with $n\ge\max(j,k)$. Its coefficient is therefore proportional to the number of links to the right of both charges. Charges that are far apart can interact in the reduced Hamiltonian. Explicit-link locality has been exchanged for a smaller constrained register.

4. Solve the two-site benchmark

Diagonalize

$$H = \begin{pmatrix} -m&-w\\ -w&m+\kappa/2 \end{pmatrix}$$

and derive the pair-state probability from the initial vacuum vector $(1,0)^{\mathsf T}$.

Solution

Write

$$H = \frac\kappa4I -wX -\Delta Z, \qquad \Delta=m+\frac\kappa4.$$

The eigenvalues are

$$E_\pm = \frac\kappa4 \pm\sqrt{w^2+\Delta^2}.$$

Let $\Omega=\sqrt{w^2+\Delta^2}$. The irrelevant constant contributes a global phase, while

$$e^{-it(-wX-\Delta Z)} = \cos(\Omega t)I +i\frac{wX+\Delta Z}{\Omega} \sin(\Omega t).$$

The transition amplitude to the second basis state is $iw\sin(\Omega t)/\Omega$, so

$$P_{\mathrm p}(t) = \frac{w^2}{w^2+\Delta^2} \sin^2(\Omega t).$$

5. Compare two finite link models

For basis states $\lvert-1\rangle,\lvert0\rangle,\lvert1\rangle$, compare a hard-truncated $U(1)$ shift with a cyclic $\mathbb Z_3$ shift. Write the action on every basis state and state whether each shift is unitary. With $E=\operatorname{diag}(-1,0,1)$, also evaluate $[E,U]$ for both models.

Solution

The hard shift acts as

$$U_{\mathrm{hard}}\lvert-1\rangle=\lvert0\rangle, \qquad U_{\mathrm{hard}}\lvert0\rangle=\lvert1\rangle, \qquad U_{\mathrm{hard}}\lvert1\rangle=0.$$

It is not unitary because it annihilates a nonzero state. The cyclic shift acts as

$$U_{3}\lvert-1\rangle=\lvert0\rangle, \qquad U_{3}\lvert0\rangle=\lvert1\rangle, \qquad U_{3}\lvert1\rangle=\lvert-1\rangle.$$

It permutes an orthonormal basis and is unitary. The wraparound transition is absent in the hard rotor truncation and present in the finite group. Equal dimension does not make the models equivalent. Direct application to each basis vector gives

$$[E,U_{\mathrm{hard}}]=U_{\mathrm{hard}},\qquad [E,U_3]=U_3-3\lvert-1\rangle\langle1\rvert.$$

On the wrapped state, $[E,U_3]\lvert1\rangle=-2\lvert-1\rangle$, not $U_3\lvert1\rangle=\lvert-1\rangle$.

6. Show exact product-formula gauge preservation

Suppose $H=A+B$ with $[A,G_x]=[B,G_x]=0$ for every vertex. Prove that first- and second-order product formulas commute with every $G_x$ at any step size. Does this prove that they implement exact time evolution?

Solution

If $[A,G_x]=0$, then every power of $A$ commutes with $G_x$, so $[e^{-iA\tau},G_x]=0$. The same holds for $B$. Products of operators that commute with $G_x$ also commute with it. Therefore both

$$e^{-iA\tau}e^{-iB\tau}$$

and

$$e^{-iA\tau/2}e^{-iB\tau}e^{-iA\tau/2}$$

preserve the Gauss sector exactly for any $\tau$. They still have product-formula error when $[A,B]\ne0$. Gauge preservation proves structural validity, not exact dynamics.

7. Find the residual effect of a penalty

Let $H_0$ preserve the physical projector $P$, let $Q=I-P$, and add $\lambda Q$. A perturbation $V$ couples $P$ and $Q$. What is the scale of the leading virtual correction within $P$ when $\lambda$ is the dominant gap? Why can increasing $\lambda$ also make digital simulation harder?

Solution

Second-order elimination gives a physical-space correction schematically

$$\Delta H_{\mathrm{eff}} \simeq -PVQ \frac1{\lambda+QH_0Q-E} QVP.$$

When $\lambda$ dominates the other scales,

$$\lVert\Delta H_{\mathrm{eff}}\rVert = O\!\left( \frac{\lVert PVQ\rVert^2}{\lambda} \right).$$

Thus leakage can be suppressed while residual in-sector dynamics remains. Larger $\lambda$ increases the Hamiltonian norm and fastest frequency, which can require smaller product-formula steps, finer control, greater synthesis precision, or a larger block-encoding normalization. The useful penalty depends on both leakage suppression and implementation error.

8. Compare mean constraints with total leakage

Suppose a constraint has eigenvalues $-1,0,1$. Does $\langle G\rangle=0$ imply the state is physical? Next consider $N_v$ independent constraints, each violated with probability $p$ and otherwise zero. Find the per-vertex squared residual and the probability that at least one constraint fails.

Solution

An equal mixture of the $+1$ and $-1$ eigenstates has zero mean but $\langle G^2\rangle=1$, with no physical weight. For the independent constraints,

$$\epsilon_G=p,\qquad p_{\mathrm{out}}=1-(1-p)^{N_v}\le N_vp.$$

At $p=10^{-3}$ and $N_v=1000$, the mean local residual is small while $p_{\mathrm{out}}\simeq0.632$. Independence gives the exact expression; the upper bound follows without independence. More generally the positive constraint sum $K\ge\Delta_G(I-P)$ implies $p_{\mathrm{out}}\le\langle K\rangle/\Delta_G$.

References

• M. C. Bañuls and K. Cichy, “Review on novel methods for lattice gauge theories,” Reports on Progress in Physics 83, 024401 (2020), doi:10.1088/1361-6633/ab6311.
• C. W. Bauer et al., “Quantum simulation for high-energy physics,” PRX Quantum 4, 027001 (2023), doi:10.1103/PRXQuantum.4.027001.
• M. Dalmonte and S. Montangero, “Lattice gauge theory simulations in the quantum information era,” Contemporary Physics 57, 388–412 (2016), doi:10.1080/00107514.2016.1151199.
• Z. Davoudi, I. Raychowdhury, and A. Shaw, “Search for efficient formulations for Hamiltonian simulation of non-Abelian lattice gauge theories,” Physical Review D 104, 074505 (2021), doi:10.1103/PhysRevD.104.074505.
• Z. Davoudi, A. F. Shaw, and J. R. Stryker, “General quantum algorithms for Hamiltonian simulation with applications to a non-Abelian lattice gauge theory,” Quantum 7, 1213 (2023), doi:10.22331/q-2023-12-21-1213.
• J. C. Halimeh, H. Lang, J. Mildenberger, Z. Jiang, and P. Hauke, “Gauge-symmetry protection using single-body terms,” PRX Quantum 2, 040311 (2021), doi:10.1103/PRXQuantum.2.040311.
• J. B. Kogut, “An introduction to lattice gauge theory and spin systems,” Reviews of Modern Physics 51, 659–713 (1979), doi:10.1103/RevModPhys.51.659.
• J. Kogut and L. Susskind, “Hamiltonian formulation of Wilson’s lattice gauge theories,” Physical Review D 11, 395–408 (1975), doi:10.1103/PhysRevD.11.395.
• C. Kokail et al., “Self-verifying variational quantum simulation of lattice models,” Nature 569, 355–360 (2019), doi:10.1038/s41586-019-1177-4.
• E. A. Martinez et al., “Real-time dynamics of lattice gauge theories with a few-qubit quantum computer,” Nature 534, 516–519 (2016), doi:10.1038/nature18318.
• A. F. Shaw et al., “Quantum algorithms for simulating the lattice Schwinger model,” Quantum 4, 306 (2020), doi:10.22331/q-2020-08-10-306.
• Y. Tong, V. V. Albert, J. R. McClean, J. Preskill, and Y. Su, “Provably accurate simulation of gauge theories and bosonic systems,” Quantum 6, 816 (2022), doi:10.22331/q-2022-09-22-816.
• U.-J. Wiese, “Ultracold quantum gases and lattice systems: quantum simulation of lattice gauge theories,” Annalen der Physik 525, 777–796 (2013), doi:10.1002/andp.201300104.
• E. Zohar, J. I. Cirac, and B. Reznik, “Quantum simulations of lattice gauge theories using ultracold atoms in optical lattices,” Reports on Progress in Physics 79, 014401 (2016), doi:10.1088/0034-4885/79/1/014401.
• E. Zohar, A. Farace, B. Reznik, and J. I. Cirac, “Digital lattice gauge theories,” Physical Review A 95, 023604 (2017), doi:10.1103/PhysRevA.95.023604.