Gauss's law on a lattice.
Which flux patterns a gauge theory allows, and how to label them
In this exploration
Gauge theory on the lattice
What is a gauge theory? One way to define it is by a constraint. Start from a large space of states, then keep only the states that obey a local rule at every point of space. If you have learned about electromagnetism, you will recall that the divergence of the electric field, , must be directly proportional to the charge density at that point. That is a special case of a general relation in gauge theories called Gauss’s law, which relates the electric flux leaving any small region to the charge inside it. The physical states are the ones that satisfy Gauss’s law everywhere.
Why not start directly in the smaller space of physical states? The larger space is built from independent pieces, one for each edge of the grid used below, so local quantities are easy to write down: the field on an edge, the energy it costs, or a move that changes a few edges at once. It is also clear what “nearby” means: two states are close when they differ on a few edges. The physical states do not split into independent local pieces, because the rule ties neighboring edges together. Working among them directly is awkward, so it is usually simpler to keep the constraint as a rule that we check afterward.
In this article I focus on lattice gauge theories, where space is a grid and the constraint can be checked at each vertex, a point where edges of the grid meet. The figure below lets you do exactly that.
Draw a row of squares on graph paper and let every edge, or link, either carry a unit of electric flux or carry nothing. Four squares already have thirteen links and 8,192 ways to fill them. Which of these patterns can a gauge theory actually admit?
The answer depends on the choice of Gauss’s law, imposed separately at every vertex. In the interactive figure, you can choose either an SU(2) Gauss’s law1 or a Gauss’s law. Click links to place flux, watch which vertices obey the rule, and see how few patterns survive as you add squares.
Flux on a chain of squares
2 squares · 4 of 7 links carry flux · Gauss’s law holds at 6 of 6 vertices. This is allowed state 2 of 4.
Flux runs around square 1. Each vertex it passes has two flux links.
Allowed: 4 of 128 patterns (2² of 2⁷)
Z₂ rule: each link is empty or carries one unit of flux. Gauss’s law holds at a vertex when an even number of flux links meet there.
Click or tap a link to add or remove flux; on a keyboard, Tab to a link and press Enter or Space. The number beside each vertex counts the flux links that meet there. Numbers inside the squares name them for the flip buttons. The electric term uses the paper’s convention, one unit per flux link, and appears only for allowed patterns.
Flux on links
A lattice gauge theory replaces continuous space by a grid. The points of the grid are vertices, the edges joining neighboring vertices are links, and the smallest closed loops of links are squares, which physicists call plaquettes. The grid in the figure is an open chain of squares one square wide, like a ladder lying on its side, whose vertical links are the rungs. Its two ends are not joined. With squares it has links and vertices.
On this grid the electric field lives on the links. Instead of an arrow at every point of space, each link records how much electric flux runs along it. Charges, if there were any, would sit on the vertices; here there are none. The four corner vertices touch two links each, and every other vertex touches three.
Gauss’s law without charges
In ordinary electromagnetism, Gauss’s law says that the flux leaving a region is fixed by the charge inside it. Where there is no charge, field lines cannot start or stop. On a lattice the smallest region is a single vertex, so the flux arriving at each vertex must equal the flux leaving it.
I will start with the simplest version, a gauge theory. Here a link carries either one unit of flux or none, and flux has no direction.2 Gauss’s law then becomes a parity rule: every vertex must touch an even number of flux links. On the chain that means 0 or 2. In other words, flux must form closed lines that never end, which is the lattice version of the rule about field lines. This step is my own warm-up.3
How many patterns survive? With squares, exactly of the patterns are allowed. That is 2 of 16 for one square, 4 of 128 for two, 8 of 1,024 for three, and 16 of 8,192 for four.
The reason is that each square offers one free choice. Flux running around the boundary of a single square is a closed loop, so it satisfies every vertex. Any combination of such loops is allowed too. A rung shared by two chosen squares receives two units of flux and ends up empty, so loops around neighboring squares merge into one loop around the rectangle they form. On this chain every allowed pattern arises in this way from exactly one set of squares, and two options per square give patterns. One consequence is easy to check in the figure: in every allowed pattern, the top and bottom links of each square agree.
The mathematics: counting allowed patterns
Write a flux pattern as one number per link , with addition modulo 2. Gauss’s law at a vertex is then a linear equation,
where the sum runs over the links touching . With links and vertices, there are such equations, but they are not independent. Every link touches exactly two vertices, so the sum of all equations reads , which holds automatically. On a connected lattice this is the only relation. The allowed patterns therefore form a space of dimension over the two-element field, and there are of them. This number is also the count of independent loops in a connected graph. For the chain, .
The square loops are a basis of this space. They are independent, because the top link of square belongs to no other square, so no combination of the other squares can produce or cancel it. Since there are of them in a -dimensional space, every allowed pattern is a unique combination of square loops. The same observation explains why top and bottom links agree: the top and bottom links of square each belong only to square , so both carry flux exactly when square is chosen.
Moves that keep the law
Take an allowed pattern and flip all four links of one square, so that flux becomes empty and empty becomes flux. I claim that this move never breaks Gauss’s law. Each vertex touches either none of that square’s links or exactly two of them. Flipping two links changes the number of flux links at a vertex by , , or , so an even number stays even.
The flips also reach everything. Starting from the empty chain, flipping a set of squares, in any order, produces exactly the pattern built from those squares’ loops. Every allowed pattern is therefore reachable from the empty chain, and in exactly one way. The flip buttons in the figure perform these moves.
In quantum mechanics, the Hamiltonian is the operator that assigns energies. In a paper with Shailesh Chandrasekharan and Tanmoy Bhattacharya, we wrote one as
where the first sum runs over links and the second over squares. The electric term costs one unit of energy for each link that carries flux; the figure’s flux count is this energy. The plaquette term flips the flux around square , with the same strength for every square.4 For positive , every off-diagonal entry of this Hamiltonian in the paper’s basis is zero or negative, which is why its Monte Carlo simulations have no sign problem.5
When flux carries spin
The gauge theories of particle physics use richer symmetry groups than . One of the simplest is SU(2), the group that rotates a spin-½ object, and in an SU(2) gauge theory the flux itself carries spin. To describe it I need one paragraph of quantum mechanics.
A quantum state is a vector, and states can be added: a superposition of two states is another state. A spin-½ object has two basis states, up and down. Two spin-½ objects have four, and among their superpositions exactly one has total spin zero. It is called the singlet, and it looks the same however the two spins are rotated together. An odd number of spin-½ objects can never have total spin zero.
In the SU(2) model of our paper, each link is either empty (spin ) or carries flux (). A flux link has a spin-½ end at each of its two vertices.6 Gauss’s law now says that the flux ends meeting at each vertex must together form a singlet. With no ends there is one way, trivially. With one or three ends there is none, and with two ends there is exactly one. The even-flux rule, which I imposed by hand in the theory, now follows from the spin. On the chain, the SU(2) theory therefore allows the same flux patterns as , each with exactly one state, and the two counts are equal. In SU(2) mode the figure draws an arc at each vertex where two flux ends are joined into a singlet.7
The paper names these pieces. The object on each link, empty or flux, is a dimer tensor, an array of numbers with one index for each of the link’s two ends. The object on each vertex, which would describe a charge there, is a monomer tensor; with no charges, every monomer is trivial. The singlets that Gauss’s law keeps at a vertex are labelled by an index .8 A basis state is fixed by three lists: the dimer on every link, the monomer on every vertex, and the singlet label at every vertex. This labelling is the monomer-dimer tensor-network basis. The arcs in the figure are singlets, not dimers; the dimers are the links.
Where SU(2) differs
The chain hides something, because none of its vertices has more than three links. A vertex of the ordinary square lattice has four. Suppose all four carry flux. Their four spin-½ ends can be split into two singlet pairs in three ways: 1 with 2 and 3 with 4, 1 with 3 and 2 with 4, or 1 with 4 and 2 with 3.
Three pictures, but only two independent states: any one pairing is a superposition of the other two. The three are not perpendicular either: any two overlap by one half.9 Here Gauss’s law leaves a genuine choice even after the flux is fixed, and the label records it. This is where SU(2) departs from . The number of singlets grows quickly with the number of ends; six spin-½ ends already have five. This is one reason the paper works on lattices with few links per vertex: in spatial dimensions, such as the honeycomb lattice with three.
To see the difference in a lattice rather than at a lone vertex, take the smallest piece of the square lattice with an interior vertex: four squares arranged two by two. The vertex at the center touches four links.
Flux on a grid of four squares
8 of 12 links carry flux · Gauss’s law holds at 9 of 9 vertices. This is allowed pattern 10 of 16, with 2 singlet states at the center.
Flux runs around squares 1 and 4, which touch only at the center. Four flux ends meet there.
Four flux ends meet at the center. Three pictures, two independent states: any two of them span the choice. Gauss’s law no longer fixes the state from the flux pattern alone, so SU(2) has 18 states where Z₂ has 16.
Allowed patterns: 16 of 4,096 · gauge-invariant states: 18
SU(2) rule: an empty link has j = 0; a flux link has j = ½ and brings a spin-½ end to each of its two vertices. Gauss’s law holds at a vertex when the flux ends there combine into a singlet. Two ends make exactly one singlet, drawn as a pale arc. The center is the only vertex where four links meet, and four ends can make two independent singlets.
Click or tap a link to add or remove flux; on a keyboard, Tab to a link and press Enter or Space. The number beside each vertex counts the flux links that meet there. Numbers inside the squares name them for the flip buttons. The center’s links are numbered 1 left, 2 up, 3 right, 4 down in the pictures.
The grid has 12 links and 9 vertices, and so 4,096 flux patterns. Both Gauss’s laws allow the same 16 of them, , as the counting rule predicts.10 In two of the 16, all four links at the center carry flux. These are the checkerboards, in which the loops around two diagonally opposite squares touch at the center. In the other 14, every vertex touches at most two flux links, and SU(2) gives one state per pattern, as on the chain. At each checkerboard, SU(2) keeps two independent singlets at the center, so it has 18 gauge-invariant states where has 16.11
Can you find the two checkerboards with the Flip square buttons? In SU(2) mode, the figure then lets you switch between the three pictures at the center. The flux pattern is the same in all three; only the way the four ends are joined changes, and any two of the pictures already give both independent states.12
From a few states to a continuum
The conventional Hamiltonian lattice gauge theory gives each link infinitely many states.13 Models like the one above are qubit regularizations: each link holds a small, fixed number of states, and the hope is to recover a continuum gauge theory by tuning the couplings. Can so few states do the job? In our paper, we showed that on a periodic SU(2) chain the string tension, the energy per unit length of a flux line stretched between two charges, decreases toward zero as grows. In two later papers with the same collaborators, we showed that plaquette chains for SU(2) and SU(3) have continuum limits with massive particles. These particles are relativistic: their energies and momenta obey relativistic dispersion relations, with a single limiting velocity common to all of them. The lattice has no built-in speed of light, so recovering one is part of what it means to reach a continuum limit.
What the figures do and do not show
Both figures, the chain and the 2×2 grid, are open and have no charges. The paper’s chain is periodic, with its ends joined into a ring, and it carries two heavy charges. On a ring, flux can also wind around the circle, which gives a second sector of states in which the top and bottom links of a square disagree. The open chain contains only the sector in which they agree. The 2×2 grid does not appear in the paper.
The flux count is only the electric term of the Hamiltonian. The paper studied its Hamiltonian on the periodic chain; I have not computed energies on the open chain or the 2×2 grid, and neither figure shows any. Neither figure shows singlet signs, which have no physical meaning.
The counts were checked by listing every flux pattern for one to four squares and for the 2×2 grid, and the SU(2) counts by building the singlet spaces directly.
Further reading
- S. Chandrasekharan, R. X. Siew, and T. Bhattacharya, “Monomer-dimer tensor-network basis for qubit-regularized lattice gauge theories”, Physical Review D 111, 114502 (2025), arXiv:2502.14175. It introduces the basis used here, the Hamiltonian without a sign problem, and the SU(2) plaquette chain.
- R. X. Siew, S. Chandrasekharan, and T. Bhattacharya, “Asymptotic freedom and massive glueballs in a qubit-regularized SU(2) gauge theory”, Physical Review D 113, 114522 (2026), arXiv:2512.11068. It connects an SU(2) plaquette chain to the quantum Ising model and finds a continuum limit with massive excitations, a one-dimensional analogue of glueballs (massive bound states of the gauge field).
- R. X. Siew, S. Chandrasekharan, and T. Bhattacharya, “Continuum limit of a qubit-regularized SU(3) lattice gauge theory on a plaquette chain”, Physical Review D 114, L071504 (2026), arXiv:2603.01215. It finds a continuum limit with massive particles for an SU(3) plaquette chain, through a connection to the three-state quantum clock model.
Footnotes
-
Admitting only the trivial or the doublet dimers on each link. Dimer is the paper’s name for the state on a link, defined later in this article; the trivial dimer is an empty link and the doublet dimer is a flux link. ↩
-
In a theory, flux adds modulo 2: two units of flux on one link count as no flux at all. One unit is therefore its own opposite, which is why flux has no direction. ↩
-
The paper I discuss below does not formulate a gauge theory and mentions only in passing. The figure’s mode is therefore not part of the paper. ↩
-
The paper’s model is SU(2), the subject of the next section, but on this chain its plaquette term changes the flux exactly as these flips do. ↩
-
A sign problem is a cancellation between positive and negative weights in a Monte Carlo sum, which can make such simulations impractical. ↩
-
A flux link therefore has states, one for each choice of up or down at its two ends. ↩
-
A singlet’s overall sign depends on conventions, such as the order in which its two ends are written. Those signs have no physical meaning, and the figure shows none. ↩
-
On the chain, takes only one value, so the flux pattern alone fixes the state. ↩
-
That is, the overlap has size one half. Its sign depends on the sign convention chosen for each singlet. ↩
-
With links and vertices, the argument in the counting box gives patterns, and the four square loops are again a basis. SU(2) allows the same patterns because an odd number of spin-½ ends cannot form a singlet. I checked the 16 by listing all 4,096 patterns, and the 18 by building the singlet spaces directly. ↩
-
The 2×2 grid is my own example, and so are its counts. The paper works on the honeycomb lattice in , the diamond lattice in , and the periodic chain. ↩
-
The label at the center takes two values, which name two perpendicular states. No two of the three pictures are perpendicular, so none of them should be read as or . ↩
-
This is the formulation of John Kogut and Leonard Susskind. The plaquette term of our paper is not theirs; it is defined directly in the monomer-dimer basis. ↩