Do Toric Duality Invariants Survive Symmetry Breaking?

Many different four-dimensional supersymmetric gauge theories can describe the same low-energy physics. A standard example comes from D3-branes (three-dimensional membranes in string theory) sitting at a singular point of a six-dimensional Calabi–Yau space. The allowed theories are "phases" related by Seiberg duality, a swap between ultraviolet descriptions that leaves the long-distance physics unchanged. These theories are drawn as dimers: bipartite graphs on a torus whose faces are gauge groups and whose edges are matter fields. INVARIANTS OF TORIC SEIBERG DUALITY showed that if the graph is drawn "isoradially", with every face inscribed in a circle of the same radius, then each edge's angle encodes that field's R-charge, the conserved charge fixed by a-maximization. The shape of the torus, summarized by its modular $j$-invariant, is then unchanged by every duality move. So are the number field generated by the R-charges and the rank of their span over the rationals. These are exact, number-theoretic labels for a whole family of dual theories.

That paper studies duality at a fixed singularity. It never asks what happens when the theory is deformed. The most basic deformation is spontaneous symmetry breaking: a matter field gets a vacuum expectation value (VEV), which in the dimer removes an edge. That flows the theory to a daughter theory at a simpler singularity and discards the parent's R-symmetry in favor of a new one. This note proposes to map how the duality invariants move along these Higgsing flows. The main payoff is using well-behaved parents to fix the daughter data that the source paper leaves ambiguous in non-minimal phases such as $L^{2,2,2}$.

Core Claim

We conjecture that Higgsing commutes with toric Seiberg duality at the level of invariants. Fix a parent geometry. Take any two dual phases, Higgs a field in each that corresponds to the same change of the toric diagram, and ask whether the daughter invariants agree. We claim the resulting daughter theories share the same $j(\tau_R)$, number field $\mathbb{K}$, degree $[\mathbb{K}:\mathbb{Q}]$ and rational rank, even when the daughters are different, possibly non-minimal, phases.

If this holds, a phase whose isoradial embedding is degenerate can be given unambiguous invariants by inheritance from any parent phase that Higgses down to it. That directly addresses the $L^{2,2,2}$ degeneracy flagged in INVARIANTS OF TORIC SEIBERG DUALITY. If it fails, the failure pinpoints the duality-invariant data that symmetry breaking cannot see. Either outcome turns a static list of invariants into a statement about the renormalization-group flow between singularities.

The four directions play unequal roles here. Toric duality supplies the object and the invariants. Spontaneous symmetry breaking supplies the operation. Coincident branes supply the regime, with $N$ D3-branes on the stack, and a scaling check. Physical grounding of invariants is the question itself: are $j$ and $\mathbb{K}$ physical enough to be carried by a physical flow?

The link to the phenomenology in Direction 1 is conceptual, not a model-building claim. In the models where lepton number is identified with a continuous R-symmetry (“L = R” — U(1) R as the origin of leptonic ‘RPV’), the field that gets a VEV, the sneutrino, carries R-charge. Its VEV breaks the R-symmetry, and the low-energy R-assignment is reset. Our dimer Higgsing is the clean, exactly solvable version of that situation: the VEV field carries a known R-charge (an angle), and the daughter R-charges are recomputed by a-maximization. We do not claim to derive R-parity violation from dimers.

Research Program

The first step is an exact enumeration. We will list toric diagrams (lattice polygons) up to a modest number of lattice points, including the benchmark suite of the Suspended Pinched Point (SPP), the second del Pezzo surface $dP_2$ and $L^{2,2,2}$ from INVARIANTS OF TORIC SEIBERG DUALITY. For each diagram we will generate its dimer phases by urban renewal, the local graph move that implements Seiberg duality, using SageMath-style mutation tooling as in Quiver mutations, Seiberg duality, and machine learning.

For every phase and every edge whose removal is a consistent Higgsing, we record three things:

  1. the daughter dimer;
  2. the daughter's toric diagram, which is the parent with one external point removed;
  3. the daughter R-charges from a-maximization, computed in exact algebraic arithmetic rather than floating point, so that number-field degree and rank are exact.

From the R-charges we build the isoradial embedding, read off $\tau_R$, and cross-check $\tau_R$ against the toric formula in the source paper.

The second step tests commutation. For each parent geometry we form every square of the form "phase A, Higgs, daughter A′" versus "phase B, Higgs, daughter B′", where B is dual to A and both Higgsings remove the same external point. We then test whether A′ and B′ are dual phases with equal invariants. The incidence matrix $\mathcal{M}$ is a useful bookkeeping device here, since the source paper shows the dimension of its kernel equals the number of external points. Every consistent Higgsing should lower that dimension by exactly one, which is an independent structural check.

The third step is the inheritance test on degenerate phases. For $L^{2,2,2}$ and any other degenerate daughters found in the enumeration, we compare three sets of invariants:

  1. those computed directly, where the embedding may be ambiguous;
  2. those inherited from each parent that Higgses to them;
  3. those given by the toric formula.

Agreement of (2) and (3) across all parents means Higgsing resolves the degeneracy. Disagreement between parents falsifies the commutation conjecture.

The fourth step concerns stack size, where coincident-brane scaling enters. Invariants built from a-maximization are independent of the number of branes $N$ at leading order in large $N$, since the central charge scales as $N^2$ times an $N$-independent function of R-charges. A VEV proportional to the identity on the whole stack of $N$ branes realizes partial resolution. A VEV that separates the stack instead does not change the geometry at the singularity. We will verify that only the former moves the invariants, and that the ratio of parent to daughter central charge is independent of $N$ and exceeds one, as the a-theorem requires.

A stretch goal is to carry the same bookkeeping to M2-branes probing toric Calabi–Yau fourfolds, using the quiver formalism of Klebanov et al.. There, Higgsing to D2-type physics is well understood (Higgsing M2 to D2 with gravity, Ezhuthachan et al.), but no isoradial invariant is known. We treat this as exploratory.

Baselines

The do-nothing reference is to recompute each daughter's invariants from scratch using the toric formula of INVARIANTS OF TORIC SEIBERG DUALITY, with no reference to the parent. This is the strongest competitor for the numbers themselves. On non-degenerate daughters it must agree with our inherited values, and any mismatch counts as a bug. It is run on exactly the same enumerated geometries with the same exact arithmetic. Beating it does not mean producing different numbers. It means giving phase-level answers the toric formula cannot give: which daughter phase each parent phase lands in, and unambiguous invariants where the direct embedding is degenerate.

The strongest existing method for deciding duality equivalence is learned classification. The graph network of Machines and Mathematical Mutations reports 99.2% accuracy on an unseen 11-node test set and 89.1% on 20-node quivers. Quiver mutations, Seiberg duality, and machine learning reports Naive Bayes binary accuracy around 90% across multiple classes, and notes degradation on unseen mutation classes. These results are on cluster-algebra and $Q_1$–$Q_{15}$ quiver datasets, not on toric dimers. For toric Higgsing flows specifically we found no reported result on Lacuna.

For a fair comparison we will retrain a graph network of the same depth and with the same tuning budget on our daughter quivers. We then hold out entire parent geometries and ask it to decide whether daughter pairs are dual. The exact invariants beat it if they decide every pair correctly on held-out geometries, including the degenerate ones where it should fail.

The key ablation removes the isoradial ingredient. We test whether equality of daughter toric diagrams alone, without $j$ or $\mathbb{K}$, already predicts which daughter phases are dual. If it does, the invariants add nothing to the Higgsing story.

Why not just…?

Why not just recompute the daughter's invariants from its toric diagram? For generic daughters this gives the same numbers, and we use it as the correctness check. It is silent on two things: which phase a Higgsed phase becomes, and how to assign invariants where the isoradial embedding itself is ambiguous. The $L^{2,2,2}$ degeneracy is exactly such a case in INVARIANTS OF TORIC SEIBERG DUALITY. If inheritance and the toric formula always agree, including on degenerate phases, this alternative is fully adequate for the numbers. The remaining contribution would then be the phase map, and we would say so.

Why not just use toric fans and algebraic geometry, without dimers? The fan determines the daughter geometry. It does not distinguish the dual gauge-theory phases that share one geometry, and those phases are what the commutation test compares. The ablation that drops $j$ and $\mathbb{K}$ measures directly how much the dimer-level data contributes.

Why not just train a classifier on Higgsed quivers? Learned classifiers do well in distribution but, as reported in Quiver mutations, Seiberg duality, and machine learning, degrade on unseen classes. They also give no exact certificate. Our held-out-geometry experiment quantifies this rather than assuming it.

Why involve sneutrino-style symmetry breaking at all? Only as motivation for studying VEVs of R-charged fields. The proposal stands or falls on the dimer computation. No phenomenological claim depends on it, and we do not test one.

Proposed Experiments

The decisive check is the commutation rate: across all enumerated parent geometries, the fraction of Higgsing squares whose two daughters have identical $j(\tau_R)$, degree $[\mathbb{K}:\mathbb{Q}]$ and rational rank. We would declare the conjecture supported if this is 100% outside degenerate cases.

Anchors come from known values. SPP should sit at degree 2 with $\sqrt{3}$ and $dP_2$ at degree 4, as reported in INVARIANTS OF TORIC SEIBERG DUALITY. Higgsings that land on SPP must reproduce degree 2 from every parent.

The second check is inheritance on $L^{2,2,2}$ and newly found degenerate phases. We record whether all parents induce the same invariants and whether these match the toric formula.

The third is the kernel-dimension drop of $\mathcal{M}$ by exactly one per Higgsing.

The fourth is the $N$-scaling test: a parent-to-daughter central-charge ratio that is independent of $N$ and greater than one, with invariants unchanged by stack-splitting VEVs.

The last is the classifier comparison on held-out geometries described under Baselines.

A single parent geometry whose dual phases Higgs to daughters with different $j$ would falsify the central claim. We would then study that geometry as the first known case where symmetry breaking separates Seiberg-dual theories.