Argyres–Douglas Theories, Class S, and Non-Lagrangian Interfaces

Argyres–Douglas fixed points are four-dimensional $N=2$ theories in which mutually nonlocal BPS states become massless together. They are the cleanest warning that a Coulomb-branch singularity need not admit one local electric Lagrangian, and they provide a precise bridge to class-S constructions, protected observables, and duality interfaces. This page extracts scaling data from the local Seiberg–Witten geometry, builds a compact theory card for the $(A_1,A_2)$ fixed point, and separates intrinsic four-dimensional conclusions from input inherited from a six-dimensional construction.

Required background. Use singular fibers and monodromies to recognize vanishing cycles, and higher-rank Seiberg–Witten systems to interpret spectral curves and periods.

Helpful background. Duality defects and interfaces explain how a transformation of bulk data can be implemented by a codimension-one system.

Mutually nonlocal massless states

At an ordinary smooth point of a rank- $r$ Coulomb branch, a symplectic basis of charges gives special coordinates

\Pi=(a_{D,1},\ldots,a_{D,r},a^1,\ldots,a^r)^T, \qquad Z_\gamma=\gamma^T\Pi.

If one primitive charge $\gamma$ has $Z_\gamma\to0$ , a duality frame can make it purely electric. The light hypermultiplet can then be included in a local Abelian effective action. Two charges $\gamma_1$ and $\gamma_2$ can be simultaneously electric only if their Dirac pairing vanishes:

\langle\gamma_1,\gamma_2\rangle=0.

Consequently, a collision with $\langle\gamma_1,\gamma_2\rangle\ne0$ has no frame in which every massless particle is represented by a mutually local elementary field. The infrared limit may still be a perfectly consistent local QFT; what fails is the attempt to describe all of its light charged states with one weakly coupled Abelian Lagrangian. This is the defining mechanism of the original Argyres–Douglas examples Argyres and Douglas 1995, §§2–4.

Geometrically, several discriminant components meet and their vanishing cycles have nonzero intersection. The order in which their Picard–Lefschetz monodromies are traversed therefore matters. A plot of the discriminant alone is insufficient: the charge local system and its intersection form are part of the physical statement.

Scaling from the Seiberg–Witten differential

The local curve of the simplest $(A_1,A_2)$ fixed point can be written

x^2=z^3+v\,z+u, \qquad \lambda_{\mathrm{SW}}=x\,dz.

This is a local normal form, not a claim about the complete ultraviolet curve. Scale invariance requires every surviving term to be homogeneous, while the BPS mass formula requires the Seiberg–Witten differential to have mass dimension one:

2[x]=3[z], \qquad [x]+[z]=1.

Solving gives

[z]=\frac25, \qquad [x]=\frac35, \qquad [v]=\frac45, \qquad [u]=\frac65.

The parameter $u$ is the expectation value of the rank-one Coulomb-branch operator, so its scaling dimension $\Delta(u)=6/5$ obeys the interacting $N=2$ unitarity bound $\Delta>1$ . The coefficient $v$ is a relevant deformation parameter. It is not a flavor mass: an $N=2$ flavor mass has protected dimension one. Using $v$ rather than $c$ also keeps this deformation distinct from the Weyl-anomaly coefficient $c$ . The cubic discriminant,

\Delta_{\mathrm{cubic}}=-4v^3-27u^2,

is homogeneous of dimension $12/5$ and splits the cusp into ordinary singularities away from $v=u=0$ . On the real discriminant, the parametrization

v=-3t^2, \qquad u=2t^3

makes the two real branches of the displayed real slice and their common cusp explicit. Over $\mathbb C$ , $4v^3+27u^2$ is irreducible, so these are not two independent complex discriminant components. This offers a quick independent check on the assigned weights.

The figure separates the exact local calculation from the progressively stronger physical inputs needed to identify an SCFT, its protected sector, a class-S realization, and a global duality wall. Follow the numbered evidence ladder before treating any arrow as an equivalence.

The local Argyres–Douglas cusp fixes scaling weights and its discriminant exactly, while the H0 theory card, Yang–Lee protected-sector proposal, irregular class-S construction, and T[SU(2)] interface each require additional, explicitly bounded evidence. — For $x^2=z^3+vz+u$ and $\lambda_{\mathrm{SW}}=x\,dz$ , homogeneity fixes $([z],[x],[v],[u])=(2,3,4,6)/5$ and the exact cusp $-4v^3-27u^2=0$ . The remaining cards distinguish source-backed four-dimensional protected data, the proposed and Schur-index-supported $\mathcal M(2,5)$ identification, the extra boundary data in an irregular class-S construction, and the global-form choices needed to turn $T[SU(2)]$ into a specified wall. These implications and their non-implications are preserved in the structured evidence record.

The same recipe applies more generally:

isolate the singular part of the curve or Hitchin spectral equation;
retain the deformations whose scaling is being studied;
impose homogeneity and $[\lambda_{\mathrm{SW}}]=1$ ;
classify coefficients as Coulomb expectation values, masses, or relevant couplings using the $N=2$ multiplet constraints;
check that the discriminant, periods, and proposed operator dimensions scale consistently.

A local quasi-homogeneous equation by itself does not establish that a four-dimensional fixed point exists. One also needs an embedding in a consistent ultraviolet theory or another construction, a positive special-Kähler metric on the physical region, and compatible anomaly and spectrum data.

A protected theory card for the $(A₁,A₂)$ fixed point

The minimal Argyres–Douglas theory is also called $H_0$ . In the standard normalization where a free $N=2$ vector multiplet has $(a,c)=(5/24,1/6)$ , its protected-data card is as follows. The evidence column matters: a result derived from the local curve, an exact protected identity, and a strongly supported protected-sector identification are not logically interchangeable.

Datum	Value	Evidence status	Meaning and limitation
Coulomb-branch rank	$1$	local geometry	One freely generated Coulomb operator $u$
Coulomb dimension	$\Delta(u)=6/5$	local geometry	Extracted from $[\lambda_{\mathrm{SW}}]=1$
Continuous flavor symmetry	none	protected spectrum	Hence no protected dimension-one mass parameter
Conformal anomalies	$a=43/120$ , $c=11/30$	protected anomaly relations	Intrinsic four-dimensional Weyl anomalies
Higgs branch	a point	UV embedding and protected classification	No independent Higgs-branch generator is present; this is not inferred from the local curve or anomaly arithmetic alone
Associated chiral algebra	Virasoro minimal model $\mathcal M(2,5)$	proposed and strongly supported	Central charge and Schur-index character agree; this does not reconstruct the full four-dimensional operator algebra
Two-dimensional central charge	$c_{2d}=-22/5$	exact 4d/2d relation	Satisfies $c_{2d}=-12c_{4d}$

The Coulomb dimension checks the rank-one anomaly relation

2a-c=\frac{2\Delta(u)-1}{4}=\frac7{20}.

With $c=11/30$ , this gives $a=43/120$ . The four-dimensional/two-dimensional correspondence fixes $c_{2d}=-12c=-22/5$ , while the computed $H_0$ Schur index agrees with the vacuum character of the Yang–Lee Virasoro model $\mathcal M(2,5)$ Shapere and Tachikawa 2008, §§3–5; Beem et al. 2015, §§3.2 and 4.4; Córdova and Shao 2016, §§3–4. The central charge alone would not identify a vertex algebra, and an index is a protected trace rather than the complete operator algebra. Together these checks make the $\mathcal M(2,5)$ identification compelling within the Schur sector without establishing generic correlators, long multiplets, or the complete BPS spectrum in every chamber.

The pointlike Higgs-branch entry also uses more than the cusp polynomial: it is the classification inherited from the pure- $SU(3)$ embedding with no hypermultiplet flat directions and is compatible with the protected spectrum Argyres and Douglas 1995, §§2–4. It should not be counted as an independent consequence of homogeneity, the anomaly relation, or the Schur-index match.

For any non-Lagrangian theory card, record at least the Coulomb spectrum, flavor algebra and global form when known, $a$ and $c$ , flavor central charges, Higgs-branch data, protected indices or chiral algebra, line and defect data, and the construction used to define the theory. A blank entry must mean “not established,” not “absent.”

Class S as a construction of four-dimensional theories

Class S starts from the six-dimensional $(2,0)$ theory of type $\mathfrak g$ on a punctured Riemann surface $C$ , with a partial topological twist along $C$ . Conditional on accepting that six-dimensional parent theory and its specified compactification, the low-energy limit is a four-dimensional $N=2$ theory. Its Coulomb geometry is encoded by a Hitchin system on $C$ :

\det(x-\varphi(z))=0, \qquad \lambda_{\mathrm{SW}}=x\,dz,

where $\varphi$ is the Hitchin field. For $\mathfrak g=A_{N-1}$ this becomes

x^N+\phi_2(z)x^{N-2}+\cdots+\phi_N(z)=0,

and the meromorphic $k$ -differentials $\phi_k$ encode Coulomb moduli, masses, and couplings. Allowed pole behavior is fixed by the puncture type. Irregular punctures introduce additional scale-dependent data and generate many Argyres–Douglas theories, but the irregular type, leading coefficients, and admissible deformations must all be supplied. Matching only a pole order or a local polynomial does not identify the four-dimensional theory.

Complex-structure moduli of $C$ become exactly marginal couplings when the resulting four-dimensional theory is conformal. Degeneration into pairs of pants exposes weakly coupled gauge groups joined to three-punctured fixtures. Different pants decompositions give different duality frames of the same class-S construction, which is the geometric origin of many $N=2$ S-dualities Gaiotto 2012, §§2–4.

The phrase “the theory associated with $C$ ” is incomplete unless the following choices are supplied.

Construction datum	What it controls
Six-dimensional type and possible outer-automorphism twist	Gauge algebra data and twisted sectors
Oriented surface and complex structure	Marginal couplings and duality frames
Regular or irregular puncture labels	Flavor symmetry, masses, Coulomb spectrum, and relevant couplings
Boundary conditions at punctures	The actual fixture rather than only its pole orders
Polarization and discrete global data	Genuine line operators and global form
Defects inserted before compactification	Four-dimensional defects and additional operator sectors

The six-dimensional origin is a definition or construction when that parent theory and its compactification are accepted. It is not, by itself, a four-dimensional Lagrangian derivation. Conversely, anomaly matching, index identities, or equality of Hitchin systems can test a proposed identification but do not individually prove equality of every unprotected observable.

Duality walls and non-Lagrangian interfaces

A mapping-class-group transformation of $C$ acts on the four-dimensional coupling and on electric–magnetic charges. A codimension-one duality wall implements that action physically: the bulk theory on one side is written in one frame, the bulk on the other in the transformed frame, and a three-dimensional $N=2$ or $N=4$ system supplies the boundary degrees of freedom required by gauge invariance and supersymmetry. For the basic $S$ transformation of four-dimensional $N=4$ theory with Lie algebra $\mathfrak{su}(2)$ , the canonical local wall theory is $T[SU(2)]$ Gaiotto and Witten 2009, §§3–4.

That notation does not by itself specify the global wall. At the Lie-algebra level, $T[SU(2)]$ has $\mathfrak{su}(2)_H\oplus\mathfrak{su}(2)_C$ symmetry and exchanges the two factors under mirror symmetry. The centers act trivially on local gauge-invariant operators, so the faithful local-operator symmetry is naturally $SO(3)_H\times SO(3)_C$ ; coupling either side to a four-dimensional bulk depends on whether the bulk global form is $SU(2)$ or $SO(3)$ , on the discrete theta angle, and on the chosen genuine-line lattice. A statement about an $S$ wall must therefore include that global data and its action on lines, not just the name of the three-dimensional Lagrangian.

Several logically different statements are often compressed into “there is a duality interface”:

a transformation exists on the charge lattice or protected parameters;
a supersymmetric boundary condition can be written;
a three-dimensional interface theory cancels boundary variations and anomalies;
protected partition functions compose as the expected integral kernel;
line operators cross the wall with the predicted transformation;
the interface is invertible as a full quantum defect.

Each later statement is stronger. Protected kernels and line-operator actions give exact, reproducible tests, but invertibility on the full Hilbert space is additional information. A non-Lagrangian bulk theory can therefore have a well-controlled interface sector even when no ordinary bulk Lagrangian is known.

Evidence boundaries and failure modes

Mutual nonlocality is essential. Coincident massless states with zero pairwise Dirac pairing may admit one local electric description and need not define an Argyres–Douglas point.

Scaling weights need physical labels. Homogeneity can assign a number to every coefficient, but $N=2$ representation theory decides whether that coefficient is a Coulomb expectation value, mass, or coupling.

A spectral curve is not the whole theory. It determines protected low-energy information. Global form, genuine line spectrum, defect choices, and unprotected observables require extra input.

One degeneration is not every duality frame. A pants decomposition displays a weakly coupled frame only in its degeneration region. Covering the full conformal manifold requires the mapping-class-group action and its identifications.

Protected agreement has a ceiling. Matching anomalies, indices, chiral algebras, or partition functions strongly constrains a proposal, but none alone establishes equality of all correlation functions.

The strongest warranted conclusion for the minimal example is that a consistent interacting rank-one $N=2$ SCFT is reached at a mutually nonlocal singularity, with the protected data listed above. Class S supplies a broad constructive setting and a network of duality frames, while interfaces test how those frames act. Claims about a complete nonperturbative equivalence must additionally specify global data and unprotected sectors.

Exercises

For the local family $x^2=z^3+v z+u$ with $\lambda=x\,dz$ , derive all four scaling dimensions, verify that the discriminant is homogeneous, and check the real cusp parametrization $v=-3t^2$ , $u=2t^3$ .

Solution

The equations $2[x]=3[z]$ and $[x]+[z]=1$ give $[z]=2/5$ and $[x]=3/5$ . Matching $v z$ and $u$ to $x^2$ then gives $[v]=2[x]-[z]=4/5$ and $[u]=2[x]=6/5$ . Both $v^3$ and $u^2$ have dimension $12/5$ , so $-4v^3-27u^2$ is homogeneous. Substitution of $v=-3t^2$ and $u=2t^3$ gives $-4(-27t^6)-27(4t^6)=0$ ; moreover $[t]=2/5$ makes both parametrizations dimensionally consistent. The signs $t>0$ and $t<0$ trace the two real arms, but one complex parameter $t$ normalizes the entire cusp. Since the exponents $2$ and $3$ are coprime, the cusp polynomial is irreducible over $\mathbb C$ .

Show that two charges with nonzero symplectic pairing cannot both be purely electric in any duality frame.