Seifert partition functions and fibering operators

The A-twist turned the free chiral on a sphere times a circle into a finite unpaired-mode problem. What happens when that circle is itself fibered over the sphere? The answer is organized by an operator that changes the circle bundle. Understanding its weight requires more than counting supersymmetric vacua: different vacua can interfere with different phases.

Our first reproduction is deliberately small. A compact $U(1)$ Chern–Simons theory at level $k=4$, on a degree-two circle bundle over the sphere, gives $Z=(1-i)/2$ in the prescription specified below. We will derive the finite sum, recover its measure from a contour integral, and check it through a Gaussian transformation. Changing the degree to four then produces $Z=0$, even though the theory still has four isolated vacua.

This is an established reproduction with a controlled training variation. Prepare with the A-twist dossier and the Abelian sphere integral. The circle-bundle lesson supplies geometric orientation. Twisted superpotentials and Bethe equations are introduced here.

Specify the bundle and the supersymmetric background

Write $M_{g,p}$ for a principal circle bundle of integer degree $p$ over a closed genus-$g$ surface. There are no exceptional fibers in this calculation. The main example has $g=0$, positive even $p$, fiber coordinate of period $2\pi$, and metric

$$\begin{aligned} ds^2&=\eta^2+\ell^2(d\theta^2+\sin^2\theta\,d\phi^2), &\eta&=\beta(d\tau+a),\\ a_N&=\tfrac p2(1-\cos\theta)d\phi, &a_S&=-\tfrac p2(1+\cos\theta)d\phi. \end{aligned} \tag{D03.1}$$

Here $\ell,\beta>0$ are lengths. On the overlap, $\tau_S=\tau_N+p\phi$, so $\eta$ is globally defined. The integral $\int_{S^2}da/(2\pi)=p$ fixes the degree and orientation convention. Use the positive frame $(e^1,e^2,e^3)=(\eta,\ell d\theta,\ell\sin\theta d\phi)$, and put $t=\beta p/(2\ell^2)$, which has units of inverse length. Then $d\eta=t e^2\wedge e^3$.

The theory is three-dimensional N=2, with four real supercharges in flat space and gauge rank one. It has one vector multiplet, no charged matter, no FI parameter and no superpotential. We use the full N=2 rules and retain an opposite-R-charge pair on the curved space. This does not identify gauge rank with the number of supercharges.

An explicit background in the site’s original-paper convention is

$$H=\frac{it}{2},\qquad V=-t\eta,\qquad A=A_{\rm tw}-t\eta, \qquad (A_{{\rm tw},N},A_{{\rm tw},S}) =\tfrac12(\cos\theta-1,\cos\theta+1)d\phi. \tag{D03.2}$$

The north spinors are $\zeta_N=e^{-i\phi/2}(0,1)^T$ and $\widetilde\zeta_N=e^{i\phi/2}(1,0)^T$, with the south phases reversed. Their half-angle behavior is the polar spin-frame lift, as in the product example. The matrices are $\gamma^1=\sigma_3$, $\gamma^2=-\sigma_1$, $\gamma^3=-\sigma_2$. For example, the spin connection has $\omega^{23}=-t e^1/2-\cos\theta\,d\phi$. Its fiber contribution to $\nabla_1\zeta$ is $it\zeta/4$. Substitution into the Killing equation gives $5it\zeta/4$ on both sides after the $A_1=-t$ term is included. The two transverse components and the opposite-R family cancel as well.

The source review uses a shifted connection: $A^{(R)}_{\rm CK}=A-3V/2=A_{\rm tw}+t\eta/2$. This shift is essential when comparing its equation (3.59) with D03.2. One may take the graviphoton potential $C=-i\eta$, for which $V=-i\star dC$; no central-charge source couples to matter in this pure-vector example. The auxiliary scalar $H$ fixes the corresponding two-form field strength locally. No extra flat source coupling is introduced.

Globally we choose the source’s $\nu_R=0$ spin/R prescription. The R bundle is the pullback of a square root of the base canonical bundle; on $S^2$ that root has degree $-1$. In the source’s shifted flux notation, $\widetilde{\mathfrak n}_0^R=-1$ while $\mathfrak n_0^R=0$. The latter is not the actual degree of the root. For even $p$, $\nu_R=1/2$ is another possible fiber spin/R choice; it is not used here. The geometric one-form $\eta$ here follows the Seifert fiber convention, rather than redefining the original paper’s spinor bilinear.

Take positive even integer $k$ and the source’s supersymmetric gauge CS level. Set additional gauge–R, R–R and gravitational CS levels to zero. The vector is quantized with its source-prescribed gaugino contact terms, not with its determinant phase silently removed. For $U(1)$ the gaugino is gauge-neutral, so integrating it out does not shift $k$. Evenness makes the gauge quadratic form independent of a spin refinement; we still specify the supersymmetric spin/R background and contact prescription.

From a circle determinant to a vacuum equation

The scalar and fiber holonomy of the gauge multiplet combine into a dimensionless complex variable $u=-h+i\beta\sigma$, where $h=(2\pi)^{-1}\oint a_{\rm gauge}$ and $\sigma$ has units of inverse length. In the circle-reduced description $u\sim u+1$. On a nontrivial bundle this shift also changes the representative of the flux class; it is not permission to omit the flux sum.

The effective twisted superpotential $\mathcal W(u)$ packages the holomorphic response of the circle-reduced theory. It is not the original three-dimensional superpotential. Its derivative defines the gauge-flux operator $\Pi(u)=\exp(2\pi i\partial_u\mathcal W)$. A unit of gauge flux inserts this factor. Summing integer fluxes leads to the vacuum equation $\Pi(u)=1$, called the Bethe equation here. There is no spin chain being assumed.

To connect this definition with a known determinant, a charge-one chiral in the A-twist prescription has $\Pi_\Phi(u)=1/(1-x)$, $x=e^{2\pi iu}$. Integrating its logarithm gives $\mathcal W_\Phi=\operatorname{Li}_2(x)/(2\pi i)^2$, since $d\operatorname{Li}_2(x)/dx=-\log(1-x)/x$. The logarithm requires a branch locally, whereas its exponentiated derivative is single-valued. This chiral is a teaching example, not an extra field in our pure CS theory.

For the actual theory, the specified CS prescription gives

$$\mathcal W(u)=\frac{k}{2}u(u+1),\qquad \Pi(u)=(-1)^k e^{2\pi iku}=e^{2\pi iku},\qquad \hat u_a=\frac a k,\quad a=0,\ldots,k-1. \tag{D03.3}$$

The linear term matters before imposing even $k$; deleting it in an odd-level example changes the vacuum equation. There is no Weyl quotient for $U(1)$, and the $k$ roots are isolated with $\mathcal W''=k\ne0$.

The effective dilaton $\Omega$ is the curvature coupling in the reduced two-dimensional theory. It supplies an Euler-characteristic weight. Joining a handle also supplies the determinant of the Hessian of $\mathcal W$. Together these define the handle-gluing operator $\mathcal H=e^{2\pi i\Omega}\mathcal W''$. There are no charged roots or matter fields here, and the added R contact levels vanish, so $\Omega=0$ and $\mathcal H=k$.

This explains why a sphere does not simply count the $k$ roots. It carries $\mathcal H^{-1}$ at each vacuum; a torus carries $\mathcal H^0$. Consequently the product manifolds already give $Z_{\Sigma_g\times S^1}=k^g$.

Insert the change of topology

The ordinary fibering operator raises the circle-bundle degree by one. In the circle-reduced description the Kaluza–Klein momentum has an associated background gauge field. A unit of its flux produces the fiber twist. If its mass scale is $m_{\rm KK}=1/\beta$, differentiating $m_{\rm KK}\mathcal W(u)$ at fixed dimensional complex coordinate $m_{\rm KK}u$ (including its holonomy part) gives $\mathcal W-u\mathcal W'$. Thus, in our prescription,

$$\mathcal F(u)=\exp\!\left[2\pi i(\mathcal W-u\mathcal W')\right] =e^{-\pi iku^2},\qquad Z_{g,p}=\sum_a\mathcal H(\hat u_a)^{g-1}\mathcal F(\hat u_a)^p. \tag{D03.4}$$

This source formula applies to isolated Bethe vacua, with the specified spin and contact data. It is not a prescription for a continuous Coulomb branch. For the Abelian example its application becomes the explicit finite expression

$$\boxed{Z_{g,p}=k^{g-1}\sum_{a=0}^{k-1}e^{-\pi ip a^2/k}.} \tag{D03.5}$$

There is a useful gauge check before evaluating it: $\mathcal F(u+1)/\mathcal F(u)=e^{-2\pi iku}$ for even $k$, which equals one at every root. Off a root the missing factor is $\Pi^{-1}$. The change in flux representative supplies it in the unreduced expression.

For $k=4,p=2,g=0$, the four terms of D03.5 are $(1,-i,1,-i)/4$, yielding $(1-i)/2$. The number $k=4$ counts vacua, whereas $p=2$ labels the bundle. Interchanging them changes both the geometric problem and its normalization.

The absolute phase belongs to this prescription. For example, adding an allowed gravitational CS level $k_g$ multiplies $\mathcal F$ by $e^{\pi i k_g/12}$, and hence $Z_{g,p}$ by $e^{\pi i p k_g/12}$. A comparison with another sphere integral must first match R background, orientation and determinant contacts. The source’s usual round-sphere presentation uses a different R-bundle representative; setting $p=1$ in an incompletely translated expression does not settle that comparison.

Recover the measure and check the answer independently

The rank-one contour derivation of S18, equations (6.46)–(6.51), supplies a direct explanation of the inverse Hessian. Positive and negative gauge-flux series are summed on their respective convergence contours. The resulting contour encloses the simple roots of $1-\Pi$. In our example it is

$$Z_{g,p}=\sum_{a=0}^{k-1}\oint_{\hat u_a} \frac{-k^g e^{-\pi ipku^2}}{1-e^{2\pi iku}}\,du . \tag{D03.6}$$

Each contour is counterclockwise. Near $\hat u_a$, the denominator is $-2\pi ik(u-\hat u_a)$; the integral therefore contributes $k^{g-1}e^{-\pi ip a^2/k}$. The orientation and the minus sign in the integrand must be read together. This is a residue derivation from the localized flux integral, not an independent assumption that every vacuum has unit weight.

A second, mathematical check changes the number of terms from $k$ to $p$. For positive even $k,p$, quadratic Gaussian reciprocity gives

$$\frac1k\sum_{a=0}^{k-1}e^{-\pi ip a^2/k} =\frac{e^{-i\pi/4}}{\sqrt{kp}} \sum_{b=0}^{p-1}e^{\pi ik b^2/p}. \tag{D03.7}$$

Here is a regulated derivation that fixes its phase. Define $T_\epsilon=\sqrt{\epsilon/\pi}\sum_{n\in\mathbb Z}e^{-(\epsilon+i\pi p/k)n^2}$, with $\epsilon>0$. The phase is periodic in $n$ modulo $k$. Splitting $n$ into its $k$ residue classes and applying the Gaussian Poisson formula shows that $T_\epsilon$ tends to the left side. If $A=\pi^2/(k^2\epsilon)$, the error is at most $2e^{-A}/(1-e^{-3A})$, obtained by bounding the nonzero dual Gaussian modes.

Apply Poisson summation to the entire Gaussian instead. With $a_\epsilon=\epsilon+i\pi p/k$, its dual exponent separates into $i\pi k n^2/p-b_\epsilon n^2$, where $b_\epsilon=\pi^2/a_\epsilon+i\pi k/p$. The phase now has period $p$. The identity $a_\epsilon b_\epsilon=i\pi k\epsilon/p$ fixes the limiting prefactor to $e^{-i\pi/4}/\sqrt{kp}$, with square roots continued from positive real Gaussian damping. This proves D03.7 without replacing a divergent series by a finite answer.

Each term on the right also equals a regulated Fresnel integral:

$$\lim_{\epsilon\downarrow0}\int_{\mathbb R} e^{-(\epsilon+i\pi kp)s^2+2\pi ikbs}\,ds =\frac{e^{-i\pi/4}}{\sqrt{kp}}e^{\pi ikb^2/p}. \tag{D03.8}$$

This is an independent Gaussian representation of our finite result. We do not infer a new localization measure or an unrestricted contour prescription from it. For computation, shifting to the saddle and rotating onto the decaying Gaussian contour is much more reliable than truncating an oscillating real-axis integral:

import mpmath as mp
mp.mp.dps = 50
k, p = 4, 2
bethe = sum(mp.exp(-mp.j*mp.pi*p*a*a/k)
            for a in range(k)) / k
gaussian = mp.quad(lambda x: mp.exp(-mp.pi*k*p*x*x),
                   [-mp.inf, 0, mp.inf])
dual = gaussian * sum(mp.exp(mp.j*mp.pi*k*b*b/p-mp.j*mp.pi/4)
                      for b in range(p))
print(mp.nstr(bethe, 20), mp.nstr(dual, 20))
# Both: (0.5 - 0.5j)

There is also a physical normalization check at $p=0$. Quantizing compact Abelian CS theory on a spatial torus leaves two periodic holonomies. Their symplectic form has $k$ units of flux, giving a $k$-dimensional Hilbert space. Equivalently, the two holonomy operators obey $UV=e^{2\pi i/k}VU$, with one irreducible basis $|a\rangle$, $a\in\mathbb Z_k$: $U|a\rangle=e^{2\pi ia/k}|a\rangle$ and $V|a\rangle=|a+1\rangle$. The torus trace is $k$; $g$ handles give $k^g$, and the sphere gives one state. These reproduce the product-manifold values above. The product case is computed separately; it is not a continuous limit of the integer degree.

Change one input and interpret the cancellation

Hold the theory, spin/R prescription and all contacts fixed, and change $p=2$ to $p=4$. At $k=4$, the four sphere weights become $(1,-1,1,-1)/4$, so $Z_{0,4}=0$. The roots and Hessian are unchanged. A zero on this nontrivial bundle is therefore interference, not evidence that all supersymmetric vacua disappeared.

Line insertions reveal what canceled. Define the unnormalized charge-$q$ fiber insertion $Z_{g,p}[q]=k^{g-1}\sum_a e^{-\pi ip a^2/k}e^{2\pi iqa/k}$, with integer electric charge $q$ and fiber orientation fixed by $u$. At $k=4,p=2$, $Z[1]=0$ and $Z[2]=(1+i)/2$, so the normalized charge-two expectation is $i$. At $p=4$, $Z[2]=1$ while $Z[0]=0$. The insertion remains meaningful, but $Z[2]/Z[0]$ does not exist. A program must report that distinction instead of returning a very large floating-point “expectation value.”

A bounded next project and its literature status

Literature checked 20 October 2026. The ordinary and exceptional fibering-operator construction and its spin refinements are established in S18; CK19 provides the convention dictionary used here. The later Q25 treats pure CS theories with simple gauge groups and suitable global quotients, including the contact data needed in comparisons with topological surgery. Its scope is not a generic matter-coupled quotient formula. The finite example and degree variation above are training reproductions, not novelty claims.

An achievable continuation is to add one explicitly specified exceptional fiber to this same Abelian theory. First reproduce the source’s fractional-flux sum, keep its spin refinement, and compare two equivalent Seifert presentations. Agreement means the full complex numbers agree after the declared contact translation; agreement of absolute values alone is weaker. A failure is initially a branch, orientation, flux-range or counterterm diagnostic. Only after that reproduction should one propose a matter deformation, checking isolated vacua and fresh literature for that particular question.

Exercises

1. Detect a missing Hessian

A calculation sums $\mathcal F^p$ over roots with unit weight on every base. Test it at $p=0$ on a sphere and a torus, then recover the correct genus dependence from the contour formula.

Hint 1

The number of roots is $k$ on either base. The compact CS Hilbert space on a sphere has one state.

Hint 2

Expand $1-\Pi$ to first order at a root; include the $k^g$ numerator.

Solution

Unit weights give $k$ on both product manifolds. This correctly gives the torus trace but fails on $S^2\times S^1$, whose value is one. Locally, $1-\Pi=-2\pi ik(u-\hat u)+O((u-\hat u)^2)$. The counterclockwise residue of $-k^g/(1-\Pi)$ is $k^{g-1}$. Summing the $k$ roots gives $k^g$, so a sphere gives one, a torus $k$, and genus two $k^2$. The exponent $g-1$ has a physical measure origin; it is not an arbitrary normalization applied after computing the sum.

2. Separate the level, the degree and a vanishing denominator

Compute $Z_{0,2}$ at $k=2$ using both sides of D03.7. Then compute the unnormalized charge-one insertion. Does the vanishing partition function imply no ground states on a spatial torus?

Hint 1

The two root weights have opposite signs. The charge-one insertion reverses one of them.

Hint 2

For the product torus times circle, set $g=1,p=0$ in the specified theory.

Solution

The root sum is $(1+e^{-i\pi})/2=0$. The dual sum has prefactor $e^{-i\pi/4}/2$ and terms $1+e^{i\pi}=0$, independently reproducing it. The line numerator is $(1+(-1)(-1))/2=1$. Its normalization by $Z[0]$ is undefined. On $T^2\times S^1$ there is no fibering insertion and each of the two roots contributes one, so the trace is two. These are different three-manifold observables; the cancellation on $M_{0,2}$ does not remove the two torus states.

3. Test a phase without discarding it

At $k=4,p=2$, suppose another calculation gives $(1+i)/2$. Can an error test that uses only $|Z|$ detect the discrepancy? Determine the ratio, show how an added gravitational level $k_g=3$ would produce it in this background, and identify what must be checked before calling the two theories equivalent.

Hint 1

Both magnitudes are $1/\sqrt2$. Compare the complex ratio with the factor below D03.5.

Hint 2

A contact term multiplies unnormalized line numerators and the partition function by the same background factor. One numerical agreement does not specify all backgrounds.

Solution

The ratio is $i$, so a magnitude-only test misses the phase. In this convention $e^{\pi i p k_g/12}=e^{i\pi/2}=i$ for $p=2,k_g=3$. Thus that explicitly added, integer-quantized supersymmetric gravitational contact term produces the stated value. It is a changed prescription, not a numerical correction to the original answer. To identify the two computations, inspect their actual gravitational and R contact terms, spin/R backgrounds and orientations; then test the same translation at another degree and with line insertions. For example $p=6$ predicts factor $e^{3i\pi/2}=-i$, while the same contact factor cancels from a normalized line expectation whenever $Z\ne0$. Choosing a separate fitted phase for each manifold would not establish a consistent dictionary.

Next, change the global gauge group while keeping the local algebra fixed. The quotient calculation tests genuine lines, missing bundle sectors, and the genus dependence of the answer.