Topological and Holomorphic Twists

A twist changes which combination of Euclidean rotations and R symmetry acts as the rotation group. Its first payoff is algebraic: a spinorial supercharge can become a scalar differential. Its physical payoff is conditional. A topological twist has all translations trivial in $Q$-cohomology, whereas a holomorphic twist has only the anti-holomorphic translations trivial. Neither conclusion follows from the word “twist” alone; the global bundle, quantum symmetry, $Q$-square, Ward identity, and boundary conditions must all cooperate.

Required background. Generalized Killing spinors and global spin-R bundles supplies the global bundle test. $Q$-cohomology and Hodge complexes supplies the passage from a nilpotent supercharge to cohomology.

Helpful background. Two-dimensional A- and B-twists develops the dimension-specific sigma-model constructions; here they are a worked precursor, not material to repeat.

The twist datum and the global bundle

Let $H$ be the part of the Euclidean spin group preserved by the geometry, and let $G_R$ be the R-symmetry group. A twist begins with a homomorphism

$$\rho:H\longrightarrow G_R.$$

The twisted rotation group is the diagonal subgroup

$$H_{\mathrm{tw}} =\bigl\{\bigl(h,\rho(h)\bigr):h\in H\bigr\} \subset H\times G_R.$$

For an abelian $U(1)$ R factor, inversion is a homomorphism; composing $\rho$ with it is equivalent to reversing the $U(1)$ generator, and hence the signs used to label all R charges. For a nonabelian R symmetry, however, group inversion is generally an antihomomorphism, so $h\mapsto\rho(h)^{-1}$ does not by itself define an alternative twist. An equivalent convention must instead compose $\rho$ with a genuine automorphism and transform the representations consistently. What matters is the declared restriction of representations. If a field transforms in $R_L\otimes R_R$ before twisting, afterward it is decomposed under $H_{\mathrm{tw}}$. This operation changes its geometric spin assignment, not the local interactions.

Apply the same decomposition to the supercharges. A singlet of $H_{\mathrm{tw}}$ is a candidate scalar $Q$. More generally, on a complex $n$-fold one can preserve $H=U(n)$ and select a scalar under the twisted $U(n)$ while retaining holomorphic rotations. The cohomological complex is defined on the subalgebra invariant under the bosonic square:

$$\begin{aligned} \mathcal A_{\mathrm{inv}}&=\ker Q^2,\\ H_Q(\mathcal A_{\mathrm{inv}}) &=\frac{\ker(Q:\mathcal A_{\mathrm{inv}}\to\mathcal A_{\mathrm{inv}})} {\operatorname{im}(Q:\mathcal A_{\mathrm{inv}}\to\mathcal A_{\mathrm{inv}})}. \end{aligned}$$

In a gauge theory the square commonly has the equivariant form

$$Q^2 =\mathcal L_v+\delta_{\mathrm{gauge}}(\Lambda) +\delta_R(\Theta)+\delta_F(m).$$

Thus $Q$ may be nilpotent on gauge-invariant operators fixed by the residual isometry and global transformations even when it is not nilpotent on every elementary field.

The group-theory decomposition is only the local part of the construction. On a curved manifold one also needs:

• an $H$-reduction of the frame bundle, such as an orientation or complex structure;
• a spin-R structure, including any common central quotient needed to define all fields;
• an R-symmetry background whose transition functions and connection implement $\rho$ and obey flux quantization;
• a globally defined section representing $Q$;
• a quantum-mechanically valid R-current and a consistent treatment of its anomalies.

An R-symmetry ’t Hooft anomaly does not simply erase the symmetry. It means that coupling it to a background may require inflow or additional data; without such a cancellation or trivialization, the twisted theory can be relative and the claimed metric independence can acquire an anomaly. A local scalar in the supersymmetry algebra is therefore not yet a globally defined topological field theory.

The twist is the third stage of the chapter-wide chain below. Read its two branches separately: a topological twist makes every translation $Q$ exact, whereas a holomorphic twist makes only the antiholomorphic translations $Q$ exact. The solid chain continues only after the scalar charge and its global bundle have been established.

At the twist stage, $P_\mu=\{Q,G_\mu\}$ gives topological translation invariance, while $P_{\bar i}=\{Q,G_{\bar i}\}$ gives antiholomorphic exactness and leaves holomorphic dependence. These local algebraic relations do not replace the spin–R bundle, flux-quantization, anomaly, or boundary tests shown by the dashed exits. The full eight-stage diagram is schematic and not to scale; supercharge and R-charge normalizations are theory dependent.

The reflowing text equivalent of the eight-stage chain preserves every input, construction, pass condition, output, and failure exit for narrow-screen and print reading.

Donaldson–Witten twist as a representation calculation

For four-dimensional Euclidean $\mathcal N=2$ supersymmetry,

$$\operatorname{Spin}(4)=SU(2)_+\times SU(2)_-, \qquad G_R=SU(2)_R\times U(1)_r.$$

The Poincaré supercharges transform as

$$Q^I_{\alpha}:(\mathbf2,\mathbf1;\mathbf2), \qquad \widetilde Q_{\dot\alpha I}:(\mathbf1,\mathbf2;\mathbf2),$$

where the last entry is the $SU(2)_R$ representation and the unused $U(1)_r$ charge is suppressed. The Donaldson–Witten twist uses

$$SU(2)'_+=\operatorname{diag}\bigl(SU(2)_+\times SU(2)_R\bigr).$$

Then

$$\mathbf2\otimes\mathbf2=\mathbf1\oplus\mathbf3.$$

With the orientation convention $\Lambda^2_+\simeq(\mathbf3,\mathbf1)$, the left-handed supercharges yield a scalar and a self-dual two-form, while the right-handed supercharges become a one-form. The vector-multiplet fermions reorganize in the same way into $\eta$, $\chi^+_{\mu\nu}$, and $\psi_\mu$. This is the cohomological field content used in the original construction Witten 1988, §§2–3.

Here is one off-shell convention. Take Lie-algebra fields anti-Hermitian and define

$$\delta_\phi A=D_A\phi, \qquad \delta_\phi X=[X,\phi]$$

for every other adjoint field $X$. With an auxiliary self-dual two-form $H^+$,

$$\begin{aligned} QA&=\psi, & Q\psi&=D_A\phi, & Q\phi&=0,\\ Q\bar\phi&=\eta, & Q\eta&=[\bar\phi,\phi],\\ Q\chi^+&=H^+, & QH^+&=[\chi^+,\phi]. \end{aligned}$$

Applying $Q$ twice gives $Q^2=\delta_\phi$ on every displayed field. The auxiliary $H^+$ is what makes this statement off shell; eliminating it would make part of the algebra close only after using an equation of motion. Factors of $i$ and some signs change with Hermitian generators, so the transformation table and the gauge-transformation convention must always be read together.

Gauge-invariant polynomials such as

$$\mathcal O_k^{(0)}=\operatorname{Tr}\phi^k$$

are $Q$ closed. Their descendants obey

$$Q\mathcal O_k^{(p)}+d\mathcal O_k^{(p-1)}=0.$$

If $C_p$ is a closed $p$-cycle, $\int_{C_p}\mathcal O_k^{(p)}$ is $Q$ closed. If $C'_p-C_p=\partial B_{p+1}$, then

$$\int_{C'_p}\mathcal O_k^{(p)}-\int_{C_p}\mathcal O_k^{(p)} =-Q\int_{B_{p+1}}\mathcal O_k^{(p+1)}.$$

The observable therefore depends only on the homology class in $Q$-cohomology, provided the deformation does not cross another insertion, a defect, or a boundary Witten 1988, §3, pp. 370–371, Eqs. (3.34)–(3.39).

Topological and holomorphic protection

The flat-space algebra gives a quick local diagnostic:

Twist type	Translation relation	Expected position dependence
topological	$P_\mu=\{Q,G_\mu\}$ for every direction	locally constant away from collisions
holomorphic	$P_{\bar i}=\{Q,G_{\bar i}\}$, while $P_i$ may survive	holomorphic in $z^i$, with allowed singularities at collisions

For a topological claim on a curved manifold, the corresponding quantum statement is that the stress tensor is $Q$ exact up to declared anomaly or improvement terms,

$$T_{\mu\nu}=\{Q,G_{\mu\nu}\}+T^{\mathrm{anom}}_{\mu\nu}.$$

Use the Euclidean convention $\delta S=\tfrac12\int_M d^dx\,\sqrt g\,T_{\mu\nu}\delta g^{\mu\nu}$. For metric-independent insertions, a variation of a normalized correlator then has the form

$$\delta_g\langle\mathcal O_1\cdots\mathcal O_n\rangle =-\frac12\int_M d^dx\,\sqrt g\,\delta g^{\mu\nu} \langle T_{\mu\nu}\mathcal O_1\cdots\mathcal O_n\rangle_{\!c}.$$

If $T^{\mathrm{anom}}_{\mu\nu}=0$, the $Q$ Ward identity has no measure or boundary remainder, and the insertions stay separated, the right-hand side vanishes. This is the derivation; “$Q$ is a scalar” by itself is not enough Witten 1988, §2.2, p. 361, Eq. (2.33), and §3, p. 365, Eqs. (3.8)–(3.9).

For a holomorphic twist, the same argument applies only in anti-holomorphic directions. Locally,

$$\partial_{\bar z^{\bar i}}\langle\mathcal O_1(z_1)\cdots\mathcal O_n(z_n)\rangle =\langle\{Q,\cdots\}\rangle=0$$

away from contact loci, while holomorphic derivatives need not vanish. In four-dimensional $\mathcal N=1$ theories with a $U(1)_R$ symmetry, the rigid supersymmetric background on a compact four-manifold selects complex geometry, and the twisted variables make the holomorphic dependence explicit Closset et al. 2014, §§2–4. This result is not a claim that every complex manifold and every R-charge assignment supports the same theory: the global bundles, background fields, and anomalies remain part of the input.

What can obstruct the conclusion

A publication-level twist record should answer each of these questions:

1. Twist datum: What are $H$, $G_R$, $\rho$, and the representations of all fields and supercharges after restriction?
2. Global lift: Which spin, spin-c, or spin-R bundle exists, and are all R fluxes compatible with the charge lattice?
3. Closure: What is $Q^2$ on fields, gauge-invariant operators, and extended operators? Is closure off shell or only on shell?
4. Quantum symmetry: Is the R-current preserved by the regulator? If it has an anomaly, what cancels or records it?
5. Protected directions: Which translations or stress-tensor components are actually $Q$ exact?
6. Ward identity: Are the measure, contour, boundary conditions, and insertions $Q$ invariant, including contact terms?
7. Residual data: Which metric, complex-structure, bundle, counterterm, boundary, or polarization dependence remains?

A failure at an early step cannot be repaired by a later formal cohomology calculation. Conversely, a holomorphic twist is not an incomplete topological twist: retaining holomorphic dependence is its intended protected structure.

Common pitfalls

Treating the twist as a change of dynamics. The representation reassignment is kinematic. Dynamics enters when one proves the quantum Ward identities and identifies the surviving $Q$-cohomology.

Equating scalar with nilpotent. A scalar supercharge may square to a gauge transformation, isometry, or global symmetry. State the invariant subalgebra on which its cohomology is taken.

Calling every metric-independent classical action topological. A quantum anomaly, a spacetime boundary, contact terms, or local background counterterms can restore metric or framing dependence.

Exercises

1. Twisted supercharges. Decompose both chiralities of the four-dimensional $\mathcal N=2$ supercharges under $SU(2)'_+\times SU(2)_-$.

Solution

For $Q^I_\alpha$, the $SU(2)_+$ and $SU(2)_R$ doublets combine as $\mathbf2\otimes\mathbf2=\mathbf1\oplus\mathbf3$, so this chirality gives $(\mathbf1,\mathbf1)\oplus(\mathbf3,\mathbf1)$: a scalar and a self-dual two-form. For $\widetilde Q_{\dot\alpha I}$, the $SU(2)_R$ doublet becomes an $SU(2)'_+$ doublet, so the result is $(\mathbf2,\mathbf2)$, the vector representation. The identification of the triplet with self-dual rather than anti-self-dual forms uses the stated orientation convention.

2. Homological invariance. Let $C'_p-C_p=\partial B_{p+1}$. Derive the difference between the two integrated descendants.

Solution

Stokes’ theorem and the $(p+1)$-st descent equation give

$$\int_{C'_p}\mathcal O^{(p)}-\int_{C_p}\mathcal O^{(p)} =\int_{B_{p+1}}d\mathcal O^{(p)} =-Q\int_{B_{p+1}}\mathcal O^{(p+1)}.$$

The two insertions define the same $Q$-cohomology class. If $B_{p+1}$ crosses another operator or meets a boundary, extra contact or endpoint terms must be included.

3. Holomorphic versus topological. Suppose $P_{\bar z}=\{Q,G_{\bar z}\}$ but $P_z$ is not $Q$ exact. What can be concluded about a separated two-point function of $Q$-closed operators?

Solution

The Ward identity makes its $\bar z$ derivative vanish, so the correlator is holomorphic away from collisions. Nothing forces the $z$ derivative to vanish. Poles or other holomorphic singularities at coincident points are compatible with the conclusion because the separated-insertion Ward identity does not control contact terms.

References

• Closset, Cyril, Thomas T. Dumitrescu, Guido Festuccia, and Zohar Komargodski. “From Rigid Supersymmetry to Twisted Holomorphic Theories.” Physical Review D 90 (2014): 085006. doi:10.1103/PhysRevD.90.085006. Open preprint.
• Witten, Edward. “Topological Quantum Field Theory.” Communications in Mathematical Physics 117 (1988): 353–386. doi:10.1007/BF01223371. Open PDF.

Next step

The rigorous finite-dimensional model for the next part of the argument is the Atiyah–Bott–Berline–Vergne fixed-point formula.