Elliptic Genera, Anomaly Data, and c-Extremization

The elliptic genus is a supersymmetric torus trace that records protected states and their flavor charges. Its modular and elliptic transformation laws are fixed by two-dimensional ’t Hooft anomalies. In a unitary $(0,2)$ superconformal theory, the same anomaly matrix determines the exact right-moving R symmetry through c-extremization. Both tools are powerful only after gauge anomalies, spin structures, compactness, accidental symmetries, and continuum contributions have been controlled.

Required background. We use the two-dimensional superalgebra and R symmetries and ’t Hooft anomaly matching. Helpful background. Effective twisted superpotentials offer a complementary vacuum-counting calculation.

The protected torus trace

Choose periodic fermion boundary conditions along the spatial circle. At a conformal fixed point, define

$$Z(\tau,z_I)= \operatorname{Tr}_{\mathcal H_R} \left[(-1)^F \mathfrak q^{L_0-c_L/24} \bar{\mathfrak q}^{\bar L_0-c_R/24} \exp\left(2\pi i\sum_Iz_IQ_I\right) \right],$$

where

$$\mathfrak q=e^{2\pi i\tau}, \qquad \bar{\mathfrak q}=e^{-2\pi i\bar\tau}.$$

For a $(2,2)$ theory, $\mathcal H_R$ is conventionally called the RR Hilbert space because both chiralities are periodic. We reserve $q=e^{-2\pi r+i\theta}$ elsewhere for a GLSM Kähler coordinate. Away from a fixed point, the cylinder Hamiltonians $H_L$ and $H_R$ replace the shifted Virasoro zero modes.

The trace convention above treats the supersymmetry used in the index as right-moving. Then, up to the normalization of the supercharge,

$$\{\bar Q_+,\bar Q_+^\dagger\} =2\left(\bar L_0-\frac{c_R}{24}\right).$$

Differentiating the trace with respect to $\bar\tau$ inserts this anticommutator. States with positive right-moving excitation energy pair with opposite fermion parity, leaving only right-moving Ramond ground states. If the spectrum is discrete and the trace converges well enough that the pairing can be performed without a boundary term,

$$\partial_{\bar\tau}Z=0.$$

This is the same pairing argument as the Witten index, refined by left-moving energy and flavor charges. It was formulated for QFT elliptic genera in Witten 1987, pp. 526–529.

Spin structures and insertions

A torus has four spin structures. Periodic spatial fermions select $\mathcal H_R$, while inserting $(-1)^F$ makes the temporal fermions periodic in the Euclidean path integral. Thus the displayed supertrace uses the periodic–periodic spin structure. Removing $(-1)^F$ or changing the spatial sector changes that structure. Spectral flow can relate the result to NS-sector quantities only when the R-charge lattice and anomaly permit the required large transformation. Flavor holonomies $z_I$ are also periodic only up to anomaly phases.

Before quoting a genus, state:

• which chirality supplies the supercharge;
• the spatial and temporal spin structures;
• whether $F$ is total, left, or right fermion number;
• the R and flavor currents inserted;
• the charge lattice and periodicity of every $z_I$;
• the infrared prescription if the spectrum is not discrete.

Changing any of these can produce a different theta-function expression rather than a harmless notation change.

Convergence, fugacity poles, and holomorphy

The Hamiltonian trace is initially defined for $\operatorname{Im}\tau>0$ and in a chamber of complex flavor fugacities where the charge sums converge. A compact theory with finite degeneracy at each left-moving level gives the expected holomorphic genus. Meromorphy in $z_I$ signals an additional issue, such as an equivariantly regulated noncompact direction, an infinite charge tower, or poles that have not yet canceled in a gauge-residue sum.

A pole in a flavor fugacity is not automatically a modular anomaly. It often means that a bosonic zero mode or an infinite tower of charged states is no longer suppressed. The meromorphic continuation away from the convergence chamber remains useful, but its value at the pole is not a convergent trace. An equivariant flavor holonomy can regulate a noncompact direction even when the unrefined limit $z_I\to0$ diverges.

This is distinct from continuum nonholomorphy. If scattering states reach threshold, their bosonic and fermionic spectral densities need not cancel pointwise. The boundary term then leaves genuine $\bar\tau$ dependence, and modular covariance can require a nonholomorphic completion rather than a holomorphic genus Troost 2010, §§2–4 (PDF).

Anomaly matrix and Jacobi transformations

For conserved Abelian currents $J_I$, define our anomaly convention by

$$k^{IJ}=\operatorname{Tr}_{\text{Weyl fermions}} \gamma^3Q^IQ^J,$$

where $\gamma^3=+1$ on right-moving fermions and $-1$ on left-moving fermions. The gravitational anomaly is

$$k_{\mathrm{grav}}=\operatorname{Tr}\gamma^3=c_R-c_L.$$

These coefficients are RG invariant as long as the symmetries are preserved. With the holonomy convention in the trace above, the $(0,2)$ genus transforms as

$$Z\!\left(\frac{a\tau+b}{c\tau+d}, \frac{z_I}{c\tau+d}\right) =\epsilon_\gamma \exp\left[ -\frac{\pi ic}{c\tau+d}k^{IJ}z_Iz_J \right]Z(\tau,z_I),$$

where $\epsilon_\gamma$ contains the gravitational-anomaly and spin-structure multiplier. The displayed sign follows the declared $\gamma^3$ and theta-function conventions; reversing the chirality convention changes it. Under $z_I\mapsto z_I+\lambda_I\tau+\mu_I$, another anomaly-determined phase appears. In the conventional Jacobi transformation $\exp[2\pi i c\,m^{IJ}z_Iz_J/(c\tau+d)]$, the index matrix in this convention is therefore

$$m^{IJ}=-\frac12k^{IJ},$$

not $k^{IJ}$ itself. The anomaly fixes the index; the factor and sign depend on the declared Jacobi and chirality conventions Benini et al. 2015, §2.2, eqs. (2.16)–(2.18) (PDF).

Calling $Z$ a weak Jacobi form requires more than modular-looking notation. The charges must define an integral lattice or a specified multiplier system; the Fourier coefficients must be holomorphic and have the required lower bound; and poles in $z_I$ must be absent. A meromorphic refinement, a vector-valued genus, or a nonholomorphic completion should be named as such.

A Landau–Ginzburg example

For one $(2,2)$ chiral field with quasi-homogeneous weight $\omega$ and an isolated LG fixed point, the elliptic genus contribution is

$$Z_{\mathrm{LG}}(\tau,z)= \frac{\theta_1(\tau,(1-\omega)z)} {\theta_1(\tau,\omega z)}.$$

For $W=X^{k+2}/(k+2)$, $\omega=1/(k+2)$. Using $\theta_1(\tau,z)\sim z\,\theta_1'(\tau,0)$ as $z\to0$,

$$Z_{\mathrm{LG}}(\tau,0) =\frac{1-\omega}{\omega}=k+1.$$

This equals the dimension of $\mathbb C[X]/(X^{k+1})$ and the number of vacua after a generic massive deformation. The equality is a sharp cross-check among the genus, chiral ring, and Witten index. Orbifolding requires a sum over commuting spatial and temporal twists, not merely projection of this untwisted answer; explicit LG orbifold formulas appear in Kawai, Yamada, and Yang 1994, §§2–3 (PDF).

Gauge theories and Jeffrey–Kirwan residues

On $T^2$, a gauge field can have flat holonomies $u$ valued in the complexified Cartan torus. Supersymmetric localization reduces a compact gauge-theory genus to Jeffrey–Kirwan residues of a meromorphic top form,

$$Z=\frac1{|W_G|} \sum_{[P]\,\text{flat}} \sum_{u_*} \operatorname{JK-Res}_{u=u_*} \bigl(\mathsf Q(u_*),\eta\bigr) \,Z_{\mathrm{1-loop}}(u,z)\,d^{\operatorname{rk}G}u,$$

where $W_G$ is the Weyl group, $[P]$ runs only over topological sectors that admit the required flat connection, and the residue is determined by the charge covectors $\mathsf Q(u_*)$ and a generic $\eta$. For a connected simply connected group, the usual formula has one flat-bundle sector. Chiral, Fermi, and vector multiplets contribute ratios of theta functions whose poles occur where charged modes become massless; this is a holonomy sum, not the magnetic-flux sum of a three-dimensional twisted index.

The prescription requires:

1. cancellation of local gauge anomalies, so the integrand descends to the holonomy torus;
2. a choice of global gauge group and bundle sectors;
3. consistent R charges and spin structure;
4. a treatment of nonprojective or degenerate pole arrangements;
5. control of boundary contributions from noncompact Coulomb directions.

For a projective pole arrangement on the compact holonomy torus, the complete sum is independent of the auxiliary $\eta$ even though individual residues are not. Nonprojective arrangements and Coulomb-boundary contributions require an extended prescription rather than an assumed cancellation. Benini, Eager, Hori, and Tachikawa derived the general-rank formula and tested it on Abelian and non-Abelian dualities Benini et al. 2015, §§2–4 (PDF).

c-extremization

Consider a theory that flows to a unitary $(0,2)$ SCFT. Let $R_0$ be a reference R current and let $F_I$ run over every Abelian infrared flavor current that commutes with the chosen supercharges and is eligible to mix without breaking a preserved non-Abelian symmetry. A trial R current is

$$R_{\mathrm{tr}}(t)=R_0+\sum_It_IF_I.$$

The trial right-moving central charge is

$$c_R^{\mathrm{tr}}(t)=3k^{RR}(t) =3\left(k^{R_0R_0}+2t_Ik^{R_0I}+t_It_Jk^{IJ}\right).$$

Stationarity gives

$$\frac{\partial c_R^{\mathrm{tr}}}{\partial t_I} =6k^{RI}(t)=0.$$

Thus the exact superconformal R current has zero mixed anomaly with every eligible flavor current. After solving for $t_I^*$,

$$c_R=3k^{RR}(t^*), \qquad c_L=c_R-k_{\mathrm{grav}}.$$

This is an extremization, not universally a maximization. In a normalizable unitary CFT, a purely right-moving flavor current has positive $k^{II}$ and gives a locally minimizing direction, while a purely left-moving current has negative $k^{II}$ and gives a locally maximizing direction. The theorem applies to a unitary $(0,2)$ conformal fixed point with $L_0,\bar L_0$ bounded below, a normalizable vacuum, normalizable local currents, and a complete infrared mixing family. A compact discrete spectrum is a sufficient setting, not the theorem’s minimal hypothesis. Enhanced $(2,2)$ theories may be treated after choosing a $(0,2)$ subalgebra; a $(2,0)$ convention reverses the chirality assignments. The derivation and hypotheses are in Benini and Bobev 2013, pp. 1–3, eqs. (2)–(7) (PDF).

A one-flavor calculation

With one flavor current $F$ and $k^{FF}\ne0$,

$$t_*=-\frac{k^{R_0F}}{k^{FF}},$$

and

$$c_R=3\left[ k^{R_0R_0}-\frac{(k^{R_0F})^2}{k^{FF}} \right].$$

If $k^{FF}=0$ but $k^{R_0F}\ne0$, no stationary point exists within this trial family. That is a diagnostic: a symmetry or sector is missing, the assumed fixed point may not exist, or the current is not an ordinary flavor current eligible for mixing.

One anomaly data table, three uses

A useful comparison begins by refusing to identify unlike quantities. In two dimensions, $k^{AB}$ is a dimensionless chiral ’t Hooft-anomaly coefficient: it controls current nonconservation, elliptic transformation phases, and c-extremization. In three dimensions, $\kappa_{AB}$ is a dimensionless parity-odd contact coefficient multiplying $(4\pi)^{-1}\int A_A\wedge dA_B$: its fractional background class and its exactly quantized dynamical representative constrain response. Neither coefficient can be inserted into the other dimension’s extremization rule.

The three fixtures below freeze their dimensions, global forms, periods, regulators, deformations, and evidence ceilings. The first is an exact compact Landau–Ginzburg calculation. The second is an induced-level calculation, not a claim that a remaining dynamical $U(1)_0$ field is automatically gapped. The third is a response test of a conjectured nonsupersymmetric duality. A complete structured equivalent is available as JSON.

Dimension and fixture	Algebra, fields, and normalization	Anomaly or contact datum	R current, operators, and deformation	Check and evidence ceiling
2d — the chapter’s $A_k$ LG convention in a $(0,2)$ presentation (often called $A_{k+1}$ in singularity indexing)	$\Phi$ chiral and $\Gamma$ Fermi, $E=0$, $J=\Phi^{k+1}$. Right-moving $\psi_+$ has $(R,F)=(r-1,1)$; left-moving $\gamma_-$ has $(R,F)=(1-(k+1)r,-(k+1))$.	$k^{AB}=\operatorname{Tr}\gamma^3Q^AQ^B$ is dimensionless, with $\gamma^3=+1$ right and $-1$ left. Thus $k^{FF}=-k(k+2)$, $k^{RF}=r-1+(k+1)[1-(k+1)r]$, and $k_{\mathrm{grav}}=0$. There is no gauge field. Integral $F$ charges give $z\sim z+1$ with anomaly phases under elliptic shifts.	c-extremization sets $k^{RF}=0$: $r_*=1/(k+2)$ and $c_R=c_L=3k/(k+2)$. Local chiral operators are classes in $\mathbb C[\Phi]/(\Phi^{k+1})$; lower monomials in $J$ deform the critical points.	Exact compact-LG result if the fixed point is unitary and isolated and the mixing family is complete: $m^{FF}=k(k+2)/2$ and the unrefined genus and Witten index are $k+1$. Half-integral $m^{FF}$ uses a multiplier system. No duality is asserted; accidental or nonnormalizable currents and continua are stopping conditions.
3d — $\mathcal N=2$ charge-one chiral induced-level fixture	Closed oriented spin three-manifold; compact $U(1)$ with $\int_\Sigma F/(2\pi)\in\mathbb Z$; $\operatorname{CS}[A]=(4\pi)^{-1}\int A\wedge dA$. One charge-one chiral uses $k_{\mathrm{bare}}=-1/2$.	A massive fermion shifts $\Delta k=\tfrac12\operatorname{sgn}M$, so $k_{\mathrm{eff}}=0$ for $M>0$ and $-1$ for $M<0$. Levels are dimensionless. Background integer shifts are scheme choices; dynamical quantization $k_{\mathrm{bare}}+\tfrac12\sum_iq_i^2\in\mathbb Z$ is compulsory.	The scalar has R charge $r$ and its fermion $r-1$. The sign of the real mass $M$ selects the chamber. Two-dimensional c-extremization does not apply; 3d $\mathcal N=2$ fixed points use F-maximization, while contact terms constrain response rather than determine $r$.	The regulated one-loop induced-level shift and large-gauge test are exact. A remaining dynamical $U(1)_0$ field is not automatically a gapped empty phase; its fate needs Maxwell, FI, monopole, and global data. No duality is asserted. Boundary inflow and Pin refinements require extra data.
3d — regulated Abelian bosonization seed	No superalgebra in the displayed descendant. A charge-one Dirac fermion with $-\tfrac12\operatorname{CS}[A]$ is compared with a Wilson–Fisher scalar coupled to compact $b$ through $\operatorname{CS}[b]+\operatorname{BF}[b;A]$, where $\operatorname{BF}[b;A]=(2\pi)^{-1}\int b\wedge dA$.	The fermion chambers have background levels $0$ and $-1$. The bosonic Higgs chamber has zero response; the unHiggsed $U(1)_1$ chamber induces $-\operatorname{CS}[A]$ plus its convention-matched invertible spin/framing response. Orientation and regulator choices are part of the statement.	There is no R-extremization for this nonsupersymmetric seed. $j_f^\mu\longleftrightarrow(2\pi)^{-1}\epsilon^{\mu\nu\rho}\partial_\nu b_\rho$; the fermion maps to a gauge-invariant dressed monopole. For the displayed $-r\lvert\phi\rvert^2$, $M>0\longleftrightarrow r>0$ is the Higgsed chamber.	Current variation and the two gapped response calculations are exact checks once the seed is stated. Equality of the interacting finite-rank critical points is conjectural; phase matching is necessary, not sufficient.

As an independent numerical checkpoint, choose $k=3$, so the chapter’s convention has $W=\Phi^5/5$. Directly summing the two fermions gives

$$\begin{aligned} r_*&=\frac15, & k^{FF}&=-15, & k^{RF}_*&=0,\\ k^{RR}_*&=\frac35, & c_R=c_L&=\frac95, & \mathcal I_{\mathrm W}&=4. \end{aligned}$$

The Jacobi convention above then gives $m^{FF}=-k^{FF}/2=15/2$; this half-integral index is meaningful together with its multiplier system. For a gauged two-dimensional model, the analogous fermion matrix must separately cancel every required gauge anomaly, reproduce the global Jacobi phases, and yield the R/flavor stationarity equations. A mismatch is usually a charge-convention or omitted-fermion error.

The two-dimensional calculation follows Benini and Bobev 2013, pp. 1–3, eqs. (2)–(7) (PDF) and the Landau–Ginzburg genus check of Kawai, Yamada, and Yang 1994, §§2–3 (PDF). The three-dimensional R-current distinction is from Jafferis 2012, §§1 and 4 (PDF); the contact and bosonization rows use Redlich 1984, pp. 2366–2374, Closset et al. 2012, §§2–4 (PDF), and Seiberg et al. 2016, §§2.1–2.2 (PDF).

Accidental symmetries and continuum states

Accidental symmetry. A new Abelian current can emerge only in the infrared, so the ultraviolet anomaly polynomial omits a possible mixing direction. If an independently established free sector factorizes, include its current and anomaly and avoid double counting its contribution. There is no universal instruction to subtract an operator merely because a trial charge looks suspicious; first establish the infrared sector and its normalizable current. A stationary ultraviolet polynomial is not proof that all currents were included.

Noncompact or spontaneously broken symmetry. Its vacuum or current may fail the normalizability assumptions used in the current-algebra proof, so it cannot automatically be added to the trial family.

Continuum. In a noncompact theory, boson–fermion spectral densities can differ at threshold. Then $\partial_{\bar\tau}Z$ receives a boundary contribution, and the modular object is often a nonholomorphic completion rather than a holomorphic Jacobi form. This failure of elliptic-genus holomorphy and the failure of c-extremization’s normalizability hypothesis are related warnings, but neither result can be substituted for the other.

Wall crossing. A flavored index can jump when states enter from infinity even though local anomalies remain fixed. State the chamber and regulator.

Exercises

1. Derive the unrefined $A_k$ Witten index from the theta-function ratio.

Solution

Near zero, $\theta_1(\tau,az)=a z\theta_1'(\tau,0)+O(z^3)$. Therefore the ratio tends to $(1-\omega)/\omega$. With $\omega=1/(k+2)$, this is $k+1$.

2. A theory has $k^{R_0R_0}=2$, $k^{R_0F}=1$, $k^{FF}=-2$, and $k_{\mathrm{grav}}=1$. Find the extremum and central charges.

Solution

$t_*=-1/(-2)=1/2$. Hence $k^{RR}=2-(1)^2/(-2)=5/2$, so $c_R=15/2$ and $c_L=c_R-1=13/2$. The negative $k^{FF}$ makes this direction a local maximum, but extremality—not a universal maximum principle—is the invariant statement.

3. Why does cancellation of the gauge anomaly matter for large shifts of a holonomy $u$?

Solution

Theta functions are quasiperiodic. Under a large gauge shift of $u$, their phases multiply to an exponential governed by the gauge-anomaly coefficient. Only when the relevant gauge anomalies cancel does the one-loop integrand define a single-valued meromorphic form on the gauge-holonomy torus.

4. A flavored trace converges for $|x|<1$ to $Z(x)=\sum_{n\ge0}x^n$. What happens at $x=1$, and what does the meromorphic continuation say?

Solution

Inside the convergence chamber, $Z(x)=1/(1-x)$. Its meromorphic continuation has a pole at $x=1$, where the original trace does not converge because the flavor fugacity no longer suppresses the infinite tower. The continuation records the divergence; it does not assign a finite unrefined index. In a field theory, one must identify the zero mode or asymptotic sector responsible and state an infrared prescription.

References

• Benini, F., and Bobev, N. “Exact Two-Dimensional Superconformal R-Symmetry and c-Extremization.” Physical Review Letters 110 (2013): 061601. doi:10.1103/PhysRevLett.110.061601. Open PDF.
• Benini, F., Eager, R., Hori, K., and Tachikawa, Y. “Elliptic Genera of 2d $N=2$ Gauge Theories.” Communications in Mathematical Physics 333 (2015): 1241–1286. doi:10.1007/s00220-014-2210-y. Open PDF.
• Closset, C., Dumitrescu, T. T., Festuccia, G., Komargodski, Z., and Seiberg, N. “Comments on Chern–Simons Contact Terms in Three Dimensions.” Journal of High Energy Physics 2012, no. 9 (2012): 091. doi:10.1007/JHEP09(2012)091. Open PDF.
• Jafferis, D. L. “The Exact Superconformal R-Symmetry Extremizes $Z$.” Journal of High Energy Physics 2012, no. 5 (2012): 159. doi:10.1007/JHEP05(2012)159. Open PDF.
• Kawai, T., Yamada, Y., and Yang, S.-K. “Elliptic Genera and $N=2$ Superconformal Field Theory.” Nuclear Physics B 414 (1994): 191–212. doi:10.1016/0550-3213(94)90428-6. Open PDF.
• Redlich, A. N. “Gauge Noninvariance and Parity Nonconservation of Three-Dimensional Fermions.” Physical Review D 29 (1984): 2366–2374. doi:10.1103/PhysRevD.29.2366.
• Seiberg, N., Senthil, T., Wang, C., and Witten, E. “A Duality Web in 2+1 Dimensions and Condensed Matter Physics.” Annals of Physics 374 (2016): 395–433. doi:10.1016/j.aop.2016.08.007. Open PDF.
• Troost, J. “The Non-Compact Elliptic Genus: Mock or Modular.” Journal of High Energy Physics 2010, no. 6 (2010): 104. doi:10.1007/JHEP06(2010)104. Open PDF.
• Witten, E. “Elliptic Genera and Quantum Field Theory.” Communications in Mathematical Physics 109 (1987): 525–536. doi:10.1007/BF01208956.