betaFeedback

WorksPhysics and Astronomy

Center anomalies without twists: torsion, exact degeneracies, and condensates

Abstract

Mixed center anomalies are usually exposed by 't Hooft's twisted boundary conditions in a Hamiltonian formalism. On RP3\mathbb {RP}^3, torsion makes the twist unnecessary: a center operator on the untwisted SU(2)SU(2) Hilbert space anticommutes with time reversal in Yang-Mills theory at θ=πθ=π and with the chiral symmetry of super Yang-Mills theory, so every level is at least doubly degenerate. The relevant semiclassical object is the BPST instanton, which descends to a charge-12\tfrac12 saddle on the orbifold R4/Z2\mathbb R^4/\mathbb Z_2. We complete it in two ways: applying a conformal transformation maps the orbifold to R×RP3\mathbb R\times\mathbb {RP}^3, with the saddle describing the tunneling event between the two center sectors; by resolving the orbifold singularity, we obtain the Eguchi-Hanson space, which gives the gaugino condensate at its weak-coupling value.

Introduction

Generalized global symmetries [1] and their ’t Hooft anomalies constrain the phases of gauge theories. In 4D SU(2)SU(2) Yang–Mills (YM) theory at θ=π\theta=\pi, a mixed anomaly between time reversal and the Z2\mathbb{Z}_{2} center (1-form) symmetry forbids a trivially gapped vacuum [2]. In N=1\mathcal{N}=1 super Yang–Mills (SYM) theory, a mixed anomaly between the discrete chiral symmetry Z4χ\mathbb{Z}_{4}^{\chi} and the center forces a chiral condensate [3, 4]. In Euclidean language, these anomalies require a background two-form gauge field. In Hamiltonian language, they require ’t Hooft’s twisted boundary conditions on T3\mathbb{T}^{3} [5, 6, 7, 8, 9, 10, 11], i.e., a fractional magnetic flux, as was first pointed out in [10]: without it, center-related flat connections on T3\mathbb{T}^{3} have Chern–Simons (CS) invariants that differ only by integers, and the anomaly is invisible.

In this Letter, we point out that for SU(2)SU(2) (S)YM theory on RP3=S3/Z2\mathbb{R}\mathrm{P}^{3}=S^{3}/\mathbb{Z}_{2} no twist is needed. The torsion of RP3\mathbb{R}\mathrm{P}^{3} does the job of the magnetic flux. The untwisted SU(2)SU(2) Hilbert space contains two flat sectors whose CS invariants differ by 12\frac{1}{2}, together with an operator UU that implements the center symmetry and exchanges them. We derive the following results.

(a) In SYM theory, and in YM theory at θ=π\theta=\pi, UU anticommutes with the chiral generator and with time reversal, respectively. Every energy level is at least doubly degenerate at any radius of RP3\mathbb{R}\mathrm{P}^{3}. The twisted partition function Tr⁡[Ue−βH]\operatorname{Tr}[Ue^{-\beta H}], proposed recently as a diagnostic of monopole condensation [12], vanishes identically in both theories.

(b) One semiclassical object controls the dynamics: the BPST instanton centered at the origin descends to a charge-12\frac{1}{2} self-dual configuration on the orbifold R4/Z2\mathbb{R}^{4}/\mathbb{Z}_{2} that connects the two sectors. Its tip can be completed in two ways (Figure 1). A conformal map gives the cylinder R×RP3\mathbb{R}\times\mathbb{R}\mathrm{P}^{3}, where the configuration is a tunneling event that yields Tr⁡[Ue−βH]\operatorname{Tr}[Ue^{-\beta H}] in closed form at small radius in YM theory. Resolving the tip gives the Eguchi–Hanson (EH) space [13], on which the same saddle gives the SYM gaugino condensate at its weak-coupling value.

BPST instanton on the orbifold, its conformal cylinder, and Eguchi–Hanson resolution

Figure 1. (a) The BPST instanton centered at the origin descends to a Q=12Q=\frac{1}{2} configuration on the orbifold R4/Z2\mathbb{R}^{4}/\mathbb{Z}_{2}, with asymptotic flat connection A1A_{1} on RP3\mathbb{R}P^{3}. (b) The conformal map sends the tip to τ→−∞\tau\to-\infty: on R×RP3\mathbb{R}\times\mathbb{R}P^{3} it is a tunneling event from A0A_{0} to A1A_{1}. (c) Resolving the tip gives EH space, a cap whose bolt carries CS=0\mathrm{CS}=0. Shading: charge enclosed, from 0 to 12\frac{1}{2}.

(c) For SU(N)SU(N) on the lens spaces L(p,1)=S3/ZpL(p,1)=S^{3}/\mathbb{Z}_{p}, a temporal background sees only a Z2\mathbb{Z}_{2} part of the anomaly. It is nontrivial if and only if pp and NN contain the same power of 2.

Geometry and flat sectors

Conventions follow Ref. [14] and are collected in the End Matter. The topological charge is Q=18π2∫Tr⁡F∧FQ=\frac{1}{8\pi^{2}}\int\operatorname{Tr}F\wedge F, FaF_{a} are the color Pauli matrices, and σx,y,z\sigma_{x,y,z} are the left-invariant one-forms on S3S^{3}.

Write a point of R4\mathbb{R}^{4} as x=∣x∣x^x=|x|\hat{x}, with x^\hat{x} on the angular S3≃SU(2)S^{3}\simeq SU(2). Identifying R4\mathbb{R}^{4} with the quaternions, X≡x0I2+ixiτiX\equiv x^{0}I_{2}+ix^{i}\tau_{i}, we have X†X=∣x∣2I2X^{\dagger}X=|x|^{2}I_{2}, so X=∣x∣gX=|x|g with g∈SU(2)g\in SU(2). In Euler angles (θ,φ,ψ)(\theta,\varphi,\psi), with ψ∈[0,4π)\psi\in[0,4\pi), one has g−1dg=iτaσag^{-1}dg=i\tau_{a}\sigma_{a} (23). The flat metric is ds2=d∣x∣2+∣x∣2(σx2+σy2+σz2)ds^{2}=d|x|^{2}+|x|^{2}(\sigma_{x}^{2}+\sigma_{y}^{2}+\sigma_{z}^{2}). The reflection x→−xx\to-x is X→−XX\to-X, which acts on each angular sphere as the antipodal map ψ,θ→ψ+2π,θ→π−θ\psi,\theta\to\psi+2\pi,\theta\to\pi-\theta. The quotient S3/Z2=RP3S^{3}/\mathbb{Z}_{2}=\mathbb{R}\mathrm{P}^{3} is described by the same angles with 0≤ψ<2π0\leq\psi<2\pi. The loop cc that runs ψ:0→2π\psi:0\to2\pi at fixed (θ,φ)(\theta,\varphi) joins x^\hat{x} to −x^-\hat{x}. It is open on S3S^{3}, closed on RP3\mathbb{R}\mathrm{P}^{3}, and generates π1(RP3)=Z2\pi_{1}(\mathbb{R}\mathrm{P}^{3})=\mathbb{Z}_{2}.

We now describe the classical vacua of the SU(2)SU(2) theory on RP3\mathbb{R}\mathrm{P}^{3}. Three steps are involved.

(1) One bundle, two flat connections. Because SU(2)≃S3SU(2)\simeq S^{3} has homotopy groups π0=π1=π2=0\pi_{0}=\pi_{1}=\pi_{2}=0, every SU(2)SU(2) color bundle over a three-manifold is trivial [15]. On RP3\mathbb{R}\mathrm{P}^{3} the gauge field is therefore a single, globally defined su(2)\mathfrak{su}(2)-valued one-form, with no patching and no ’t Hooft flux. The vacua are distinguished not by the bundle but by the flat connections on it [15]. A flat connection is fixed, up to gauge, by its holonomy hch_{c} around the generator cc of π1(RP3)=Z2\pi_{1}(\mathbb{R}\mathrm{P}^{3})=\mathbb{Z}_{2}. Since c2=1c^{2}=1, hc2=Ih_{c}^{2}=I. In SU(2)SU(2) the eigenvalues of hch_{c} are then equal and ±1\pm1, so hc=±Ih_{c}=\pm I. These two central elements are not conjugate, so they label two distinct sectors:

ℓ=0:A0=0,ℓ=1:A1=−ig dg−1.(1)\ell=0:\quad A_{0}=0,\qquad\ell=1:\quad A_{1}=-ig\,dg^{-1}. \tag*{(1)}

(2) Why A1A_1 is a nontrivial vacuum. On the covering S3S^3, A1A_1 is the gauge transform of A0=0A_0=0 by gg, so it is flat. Under the antipodal map g→−gg\to-g, and the sign cancels in g dg−1g\,dg^{-1}, so A1A_1 is a well-defined flat connection on RP3\mathbb{RP}^3. It is nevertheless not gauge equivalent to A0A_0 there: since g(−x)=−g(x)g(-x)=-g(x), gg is not a single-valued map RP3→SU(2)\mathbb{RP}^3\to SU(2), hence not a gauge transformation of the SU(2)SU(2) theory. This is detected gauge invariantly by the holonomy: parallel transport along cc, which runs from xx to −x-x, gives hc=g(ψ+2π)g(ψ)−1=−Ih_c=g(\psi+2\pi)g(\psi)^{-1}=-I.

(3) The Chern–Simons invariant. A direct integration (End Matter) gives

CS⁡(A1)−CS⁡(A0)=−124π2∫RP3tr⁡(g−1dg)3=12.(2)\operatorname{CS}(A_1)-\operatorname{CS}(A_0)=-\frac{1}{24\pi^2}\int_{\mathbb{RP}^3}\operatorname{tr}(g^{-1}dg)^3=\frac{1}{2}. \tag*{(2)}

The integral over S3S^3 would be 1, the degree of gg. On RP3\mathbb{RP}^3 only half of the sphere is covered. A genuine gauge transformation h:RP3→SU(2)h:\mathbb{RP}^3\to SU(2) shifts CS by 124π2∫RP3tr⁡(h−1dh)3\frac{1}{24\pi^2}\int_{\mathbb{RP}^3}\operatorname{tr}(h^{-1}dh)^3, which is the integer degree of hh. Hence CS mod 1 is gauge invariant, and its value 0 or 12\frac{1}{2} distinguishes the sectors. The fractional value is possible only because gg is not a gauge function.

This is the key difference from T3T^3. There, the flat connections related by center transformations are Abelian, for instance A=πLτ3dxA=\frac{\pi}{L}\tau_3dx with holonomy −I-I around one cycle. Such a connection has dA=0dA=0 and A∧A∧A=0A\wedge A\wedge A=0, hence CS =0=0, so a center transformation does not change CS and the anomaly is invisible. A relative CS of 12\frac{1}{2} appears only after a spatial twist is imposed [5, 10]. On RP3\mathbb{RP}^3 the torsion of H1(RP3)=Z2H_1(\mathbb{RP}^3)=\mathbb{Z}_2 supplies it. The cycle cc has order two, and its self-linking number lk⁡(c,c)=12\operatorname{lk}(c,c)=\frac{1}{2} is exactly the fractional CS of a flat connection with holonomy −I-I around it.

The center operator. The center operator is a map from Hilbert space to itself. Its building block is the function gg of Eq. (23), which, as emphasized above, is not an SU(2)SU(2)-single-valued function on RP3\mathbb{RP}^3. Its adjoint action, however, is single-valued, because Ad⁡(−g)=Ad⁡(g)\operatorname{Ad}(-g)=\operatorname{Ad}(g). Define

U:A↦gAg−1−ig dg−1,Φadj↦gΦadjg−1.(3)U:A\mapsto gAg^{-1}-ig\,dg^{-1},\qquad\Phi_{\mathrm{adj}}\mapsto g\Phi_{\mathrm{adj}}g^{-1}. \tag*{(3)}

Because the sign of gg cancels in both expressions, UU maps SU(2)SU(2) configurations on RP3\mathbb{RP}^3 to SU(2)SU(2) configurations on RP3\mathbb{RP}^3. It preserves the action and commutes with the Hamiltonian, since every field is in the adjoint. Because gg is not an allowed gauge transformation, UU is a genuine symmetry rather than a redundancy. It is the Z2\mathbb{Z}_2 center (1-form) symmetry operator supported on RP2\mathbb{RP}^2, the Poincaré dual of cc [1, 16, 17]. Up to gauge transformations, gg can be deformed to jump by −I-I only across RP2\mathbb{RP}^2, like ’t Hooft’s electric-flux operator on a two-torus of the untwisted T3T^3 [5]. Gauge invariantly, local operators and adjoint Wilson loops are invariant, while the holonomy transforms as hc↦−ghcg−1h_c\mapsto-g h_cg^{-1}, so the fundamental Wilson loop Wc=12tr⁡hcW_c=\frac{1}{2}\operatorname{tr}h_c, which crosses RP2\mathbb{RP}^2 once, obeys

UWcU−1=−Wc.UW_cU^{-1}=-W_c.

This is the defining action of a Z2\mathbb{Z}_2 1-form symmetry. Semiclassically, Wc=(−1)ℓW_c=(-1)^\ell on sector ℓ\ell, and UU exchanges the two values. Gauging UU would give the SO(3)SO(3) theory and identify the two sectors. In the SU(2)SU(2) theory, UU maps A0A_0 to A1A_1 and shifts the CS functional by a constant (End Matter):

UCS⁡U−1=CS⁡+12.(4)U\operatorname{CS}U^{-1}=\operatorname{CS}+\frac{1}{2}. \tag*{(4)}

A nontrivial SO(3)SO(3) bundle appears only in the Euclidean trace Tr⁡[U⋯ ]\operatorname{Tr}[U\cdots]. Gluing the ends of RP3×[0,β]\mathbb{RP}^3\times[0,\beta] with UU produces the mapping torus RP3×US1\mathbb{RP}^3\times_U S^1. On this four-manifold the transition function is SO(3)SO(3)-valued, so the bundle has w2≠0w_2\ne0, with a leg only along the time direction, and charge Q=12Q=\frac{1}{2}. Every spatial slice still carries the trivial SU(2)SU(2) bundle.

Anomalies and exact consequences. SU(2)SU(2) SYM theory. With one adjoint Weyl fermion, a charge-QQ configuration carries 4Q4Q zero modes, and the conserved Z4χ\mathbb{Z}_4^{\chi} generator is X=exp⁡[iπ2(Q5−4CS⁡)]X=\exp[\frac{i\pi}{2}(Q_5-4\operatorname{CS})], where Q5Q_5 is the gaugino number [10, 11]. Since UU commutes with Q5Q_5, Eq. (4) gives

XUX−1U−1=−1.(5)XUX^{-1}U^{-1}=-1. \tag*{(5)}

YMYM theory at θ\theta. Physical states are invariant under small gauge transformations (Gauss’s law) and, since large gauge transformations commute with all gauge-invariant operators, transform as Ψ[Ah]=eiθW(h)Ψ[A]\Psi[A^h]=e^{i\theta W(h)}\Psi[A] under those of degree W(h)∈ZW(h)\in\mathbb{Z}; θ\theta labels the superselection sector, whose lowest states are the θ\theta-vacua. Time reversal T maps θ→−θ\theta\to-\theta, and the operator e2πiCS⁡e^{2\pi i\operatorname{CS}}, which is invariant under large gauge transformations, shifts θ\theta by 2π2\pi. At θ=π\theta=\pi the combination T′=e2πiCS⁡TT'=e^{2\pi i\operatorname{CS}}T is a symmetry [2, 10]. Since UU commutes with T′T', Eq. (4) gives

UT′U−1T′−1=e2πi⋅12=−1.(6)UT'U^{-1}T'^{-1}=e^{2\pi i\cdot\frac{1}{2}}=-1. \tag*{(6)}

Equations (5) and (6) are the mixed chiral–center and T–center anomalies, realized as operator algebras on the untwisted SU(2)SU(2) Hilbert space of RP3\mathbb{RP}^3. They hold at every radius RR and have two exact consequences.

(a) Degeneracy. Or (T') maps U=+1U=+1 eigenstates to U=−1U=-1 eigenstates of the same energy. Every level of SYM theory, and of YM theory at θ=π\theta=\pi, is at least doubly degenerate on RP3\mathbb{RP}^3 of any size.

**(b) Vanishing lens-space partition function. ** For N=2N=2, RP3=L(2,1)\mathbb{R}P^{3}=L(2,1) and the twisted partition function of Ref. [12] is

ZU(β)≡Tr⁡[Ue−βH]Tr⁡e−βH.(7)Z_{U}(\beta)\equiv\frac{\operatorname{Tr}[Ue^{-\beta H}]}{\operatorname{Tr}e^{-\beta H}}. \tag*{(7)}

In SYM theory, cyclicity and Eq. (5) give Tr⁡[Ue−βH]=Tr⁡[XUX−1e−βH]=−Tr⁡[Ue−βH]\operatorname{Tr}[Ue^{-\beta H}]=\operatorname{Tr}[XUX^{-1}e^{-\beta H}]=-\operatorname{Tr}[Ue^{-\beta H}], and likewise with (−1)F(-1)^{F} inserted. In YM theory at θ=π\theta=\pi, Eq. (6) with the antiunitarity of T′T^{\prime} makes the trace purely imaginary, while U2=1U^{2}=1 and [U,H]=0[U,H]=0 make it real (End Matter). In both cases (see [18] for a general statement)

ZU(β)=0for all R,β.(8)Z_{U}(\beta)=0\qquad\text{for all }R,\beta. \tag*{(8)}

Ref. [12] associates ZU→const⁡≠0Z_{U}\to\operatorname{const}\ne0 at large volume with monopole condensation; Eq. (8) does not contradict this. The two confining vacua carry Z2\mathbb{Z}_{2} SPT levels k=0,1k=0,1, with ⟨U⟩k=eiπk∫12P(B)=(−1)k\langle U\rangle_{k}=e^{i\pi k\int\frac{1}{2}P(B)}=(-1)^{k} since ∫12P(B)\int\frac{1}{2}P(B) is odd on RP3×S1\mathbb{R}P^{3}\times S^{1}: each is maximally condensed, and the two cancel in the trace.

Small RP3\mathbb{R}P^{3}: tunneling and ZUZ_{U} in closed form. At RΛ≪1R\Lambda\ll1, where Λ\Lambda is the strong scale, the coupling is weak. Because Ad⁡(−I)=1\operatorname{Ad}(-I)=1, the fluctuation spectra around A0A_{0} and A1A_{1} are identical, so the two perturbative ground states ∣0⟩,∣1⟩|0\rangle,|1\rangle are degenerate to all orders. Both holonomies are central, so SU(2)SU(2) is unbroken and there are no light WW bosons; since H1(RP3;R)=0H^{1}(\mathbb{R}P^{3};\mathbb{R})=0 there are no flat directions either, and every gluon mode is gapped at ∼1/R\sim1/R.

no flat directions either, and every gluon mode is gapped at ∼1/R\sim1/R. The degeneracy is lifted by tunneling. Writing the flat metric as ds2=d∣x∣2+∣x∣2(σx2+σy2+σz2)ds^{2}=d|x|^{2}+|x|^{2}(\sigma_{x}^{2}+\sigma_{y}^{2}+\sigma_{z}^{2}) and setting ∣x∣=Reτ/R|x|=Re^{\tau/R} gives ds2=e2τ/R[dτ2+R2(σx2+σy2+σz2)]ds^{2}=e^{2\tau/R}[d\tau^{2}+R^{2}(\sigma_{x}^{2}+\sigma_{y}^{2}+\sigma_{z}^{2})]. Thus R4∖{0}\mathbb{R}^{4}\setminus\{0\} is conformally equivalent to the cylinder R×SR3\mathbb{R}\times S^{3}_{R}, with the origin at τ→−∞\tau\to-\infty. The Hodge star on two-forms is Weyl invariant, so self-dual solutions remain self-dual, with the same action and charge. Since x→−xx\to-x preserves ∣x∣|x|, it leaves τ\tau fixed and acts on each slice as the antipodal map. Instantons on the cylinder over RP3\mathbb{R}P^{3} are therefore the Z2\mathbb{Z}_{2}-invariant instantons on R4∖{0}\mathbb{R}^{4}\setminus\{0\}.

Such an instanton exists. The self-dual BPST instanton centered at the origin is, in regular gauge,

A=x2x2+ρ2A1,(9)A=\frac{x^{2}}{x^{2}+\rho^{2}}A_{1}, \tag*{(9)}

where ρ∈(0,∞)\rho\in(0,\infty) is the size modulus. It is smooth at the origin, tends to A1A_{1} on S∞3S^{3}_{\infty}, and carries Q=1Q=1 on R4\mathbb{R}^{4}. It is invariant under x→−xx\to-x with no gauge transformation, because the reflection acts on it as a central, hence trivial, constant gauge rotation (End Matter). Its four centered moduli, the size and three SO(3)≅RP3SO(3)\cong\mathbb{R}P^{3} orientations, form the cone R4/Z2\mathbb{R}^{4}/\mathbb{Z}_{2} itself.

On R×SR3\mathbb{R}\times S^{3}_{R}, Eq. (9) is the familiar Q=1Q=1 tunneling between A0A_{0} and its large gauge transform A1A_{1}. On RP3\mathbb{R}P^{3}, A1A_{1} is not pure gauge, and the same solution becomes a genuine transition between the two sectors ℓ=0,1\ell=0,1, with half the charge. Rewritten in terms of τ\tau, the one-form Eq. (9) reads

A(τ)=A11+e−2(τ−τ0)/R.(10)A(\tau)=\frac{A_{1}}{1+e^{-2(\tau-\tau_{0})/R}}. \tag*{(10)}

a self-dual tunneling event on R×RP3\mathbb{R}\times\mathbb{R}P^{3} from A0A_{0} to A1A_{1} [Figure 1(b)], with Q=12Q=\frac{1}{2} and action 4π2/g24\pi^{2}/g^{2} (a direct check of self-duality is given in the End Matter). Unlike the fractional instantons of twisted T3×R\mathbb{T}^{3}\times\mathbb{R}, this event is known in closed form2. The instanton size ρ\rho appears as the tunneling time τ0=Rln⁡(ρ/R)\tau_{0}=R\ln(\rho/R). Each event flips the sector. Applying UU to this solution gives a self-dual event from A1A_{1} to A0A_{0}, since UU exchanges the sectors and preserves the action. Reversing τ\tau turns both into anti-self-dual events with Q=−12Q=-\frac{1}{2}. In the Euclidean amplitude ⟨ℓ′∣e−TH∣ℓ⟩\langle\ell'|e^{-TH}|\ell\rangle, each event therefore carries the weight e−4π2/g2e±iθ/2e^{-4\pi^{2}/g^{2}}e^{\pm i\theta/2}, times a one-loop determinant and the measure over its four moduli. Integrating the tunneling time τ0\tau_{0} over [0,T][0,T] gives a factor TT, and the three orientations together with the determinant give a constant KK of dimension 1/R1/R. On the two-dimensional space spanned by ∣0⟩|0\rangle and ∣1⟩|1\rangle, define the phase of ∣1⟩|1\rangle by ∣1⟩≡U∣0⟩|1\rangle\equiv U|0\rangle; then U∣1⟩=U2∣0⟩=∣0⟩U|1\rangle= U^2|0\rangle= |0\rangle. Since [U,H]=0[U,H]=0, the flip amplitudes 0→10\to1 and 1→01\to0 are equal. A single flip therefore acts on this space as UU itself, i.e., as the matrix (0110)\begin{pmatrix}0&1\\1&0\end{pmatrix}. Summing over a dilute gas of nn events gives

⟨ℓ′∣e−TH∣ℓ⟩=e−E0T∑n(2KTcos⁡θ2)nn!⟨ℓ′∣Un∣ℓ⟩=e−E0T⟨ℓ′∣e2KTcos⁡θ2U∣ℓ⟩,(11)\langle\ell'|e^{-TH}|\ell\rangle=e^{-E_0T}\sum_n\frac{\left(2KT\cos\frac{\theta}{2}\right)^n}{n!}\langle\ell'|U^n|\ell\rangle =e^{-E_0T}\langle\ell'|e^{2KT\cos\frac{\theta}{2}U}|\ell\rangle, \tag*{(11)}

where E0E_0 is the common perturbative energy. Hence

Heff=E0−2Kcos⁡θ2U,K∝(RΛ)11/3R g4(1/R).(12)H_{\mathrm{eff}}=E_0-2K\cos\frac{\theta}{2}U,\qquad K\propto\frac{(R\Lambda)^{11/3}}{R\,g^4(1/R)}. \tag*{(12)}

The power (RΛ)11/3=e−4π2/g2(1/R)(R\Lambda)^{11/3}=e^{-4\pi^2/g^2(1/R)} follows from one-loop running with b0=22/3b_0=22/3, and g−4g^{-4} comes from the four bosonic collective coordinates. The gas is dilute because KR≪1KR\ll1 for RΛ≪1R\Lambda\ll1. The energy eigenstates are the UU-eigenstates ∣k⟩=(∣0⟩+(−1)k∣1⟩)/2|k\rangle=(|0\rangle+(-1)^k|1\rangle)/\sqrt{2}, with U∣k⟩=(−1)k∣k⟩U|k\rangle=(-1)^k|k\rangle and Ek=E0−(−1)k2Kcos⁡θ2E_k=E_0-(-1)^k2K\cos\frac{\theta}{2}. Since UWcU−1=−WcUW_cU^{-1}=-W_c, with Wc=(−1)ℓW_c=(-1)^\ell, diagonal on the sectors, WcW_c maps ∣k=0⟩|k=0\rangle to ∣k=1⟩|k=1\rangle, so ⟨Wc(τ)Wc(0)⟩∝e−(E1−E0)τ\langle W_c(\tau)W_c(0)\rangle\propto e^{-(E_1-E_0)\tau} and E1−E0=4Kcos⁡θ2E_1-E_0=4K\cos\frac{\theta}{2} is the energy cost of one unit of Z2\mathbb{Z}_2 electric flux along the torsion cycle cc, the RP3\mathbb{R}P^3 analogue of ’t Hooft’s electric-flux energy on T3\mathbb{T}^3 [5]. In a confining phase one expects it to grow as σπR\sigma\pi R at large RR, where πR\pi R is the length of cc; at small RR the flux is delocalized and its cost is set by tunneling. For β≫R\beta\gg R only the states ∣k=0,1⟩|k=0,1\rangle contribute, so Tr⁡[Ue−βH]=∑k(−1)ke−βEk∝sinh⁡(2βKcos⁡θ2)\operatorname{Tr}[Ue^{-\beta H}]=\sum_k(-1)^ke^{-\beta E_k}\propto\sinh(2\beta K\cos\frac{\theta}{2}) and Tr⁡e−βH∝cosh⁡(2βKcos⁡θ2)\operatorname{Tr}e^{-\beta H}\propto\cosh(2\beta K\cos\frac{\theta}{2}), giving3

ZU(β)=tanh⁡(2βKcos⁡θ2).(13)Z_U(\beta)=\tanh\left(2\beta K\cos\frac{\theta}{2}\right). \tag*{(13)}

It vanishes at θ=π\theta=\pi, as Eq. (8) requires, and the levels cross there. Under θ→θ+2π\theta\to\theta+2\pi, the labels k=0,1k=0,1 are exchanged. This is the RP3\mathbb{R}P^3 counterpart of ’t Hooft’s electric-flux energies on twisted T3\mathbb{T}^3 [7], obtained here without a twist.

A semiclassical cap: EH space. In SYM theory, the same Q=12Q=\frac{1}{2} event carries two gaugino zero modes. It therefore cannot contribute to HH, but it gives the matrix element of O=tr⁡(λλ)\mathcal{O}=\operatorname{tr}(\lambda\lambda). On the two lowest states, O\mathcal{O} has XX-charge −1-1, so it is off-diagonal in ℓ\ell, and it is local, so it commutes with UU. Hence ⟨1∣O∣0⟩=⟨0∣O∣1⟩≡C\langle1|\mathcal{O}|0\rangle=\langle0|\mathcal{O}|1\rangle\equiv C and

Oeff=CU,⟨k∣O∣k⟩=(−1)kC,(14)\mathcal{O}_{\mathrm{eff}}=CU,\qquad\langle k|\mathcal{O}|k\rangle=(-1)^kC, \tag*{(14)}

with C∝e−4π2/g2+iθ/2C\propto e^{-4\pi^2/g^2+i\theta/2}. The cylinder considered in the YM case cannot, however, fix CC. On the round R×RP3\mathbb{R}\times\mathbb{R}P^3, supersymmetry requires a U(1)RU(1)_R background that

pure SYM lacks, so CC is unprotected and depends on RR. What is needed is a geometry that keeps the Q=12Q=\frac{1}{2} saddle with its two zero modes, preserves supersymmetry without an RR-symmetry, and is asymptotically flat, so that CC becomes the flat-space condensate.

Such a geometry is obtained by completing the cone R4/Z2\mathbb{R}^4/\mathbb{Z}_2 differently (Figure 1). The cylinder pushes the tip to τ→−∞\tau\to-\infty and keeps RP3\mathbb{R}P^3 at fixed size. Resolving the tip instead replaces it by a smooth bolt and keeps the asymptotics flat. These are two fixed backgrounds carrying the same Z2\mathbb{Z}_2-invariant BPST configuration, with Q=CS⁡(A1)−CS⁡(interior)=12Q=\operatorname{CS}(A_1)-\operatorname{CS}(\text{interior})=\frac{1}{2} in both; they are not contributions to be summed4. In YM theory, the resolved geometry is not useful quantitatively: nothing protects the amplitude and the size integral diverges in the infrared, whereas on the cylinder the radius provides the cutoff.

On the cone, before resolution, the instanton (9) has Q=12Q=\frac{1}{2}, asymptotic holonomy −I-I, and centered moduli space R4/Z2\mathbb{R}^4/\mathbb{Z}_2, as found above. What is new in SYM theory are the fermion zero modes. Two of the four undotted gaugino zero modes survive. (Dotted zero modes vanish in a self-dual background [14].) The lift (L,R)=(−I2,I2)(L,R)=(-I_2,I_2) preserves the supercharges Q~α\widetilde{Q}_\alpha and acts as −1-1 on undotted spinors. The superconformal modes λα∝Fαβξβ\lambda_\alpha\propto F_{\alpha\beta}\xi^\beta, with ξ\xi constant, are even and projected out. The superconformal modes λα∝Fαβϵβ(x)\lambda_\alpha\propto F_{\alpha\beta}\epsilon^\beta(x), with the undotted conformal Killing spinor ϵββ˙(x)=xασαβ˙ϵ\epsilon_{\beta}^{\dot\beta}(x)=x_\alpha\sigma^{\alpha\dot\beta}\epsilon, are odd and survive. The cone is scale invariant, so they are generated by broken superconformal symmetries (End Matter).

The minimal smooth resolution of this cone that preserves Q~α\widetilde{Q}_\alpha is the hyperkähler Eguchi–Hanson (EH) space, with metric

ds2=f2dr2+r2[σx2+σy2+f−2σz2],f−2=1−a4r4,(15)ds^2=f^2dr^2+r^2\left[\sigma_x^2+\sigma_y^2+f^{-2}\sigma_z^2\right],\qquad f^{-2}=1-\frac{a^4}{r^4}, \tag*{(15)}

with r≥ar\geq a, ψ∈[0,2π)\psi\in[0,2\pi), and a bolt S2S^2 at r=ar=a. It supports the SU(2)SU(2) self-dual configuration

A=τ3a2r2σz,Q(r)=12(1−a4r4),(16)A=\tau_3\frac{a^2}{r^2}\sigma_z,\qquad Q(r)=\frac{1}{2}\left(1-\frac{a^4}{r^4}\right), \tag*{(16)}

the t=at=a endpoint of a family with size modulus t∈[a,∞)t\in[a,\infty) [14]. Here Q(r)Q(r) is the charge enclosed within radius rr. In the gauge regular at the bolt, the Wilson loop around the ψ\psi-circle, W(r)=exp⁡[−iπτ3(1−a2/r2)]W(r)=\exp[-i\pi\tau_3(1-a^2/r^2)], runs from II at the bolt to −I-I at infinity. By Bianchi’s theorem, EH is a cap that prepares a state on RP3\mathbb{R}P^3 Figure 1(c) [22]: its inner slice degenerates to the bolt, which supplies CS⁡=0\operatorname{CS}=0, so Q=CS⁡(A1)−0=12Q=\operatorname{CS}(A_1)-0=\frac{1}{2}. The moduli space is a copy of EH itself [23, 24, 25]: at t=at=a the orientation orbit collapses to SO(3)/U(1)=S2SO(3)/U(1)=S^2, the bolt. The parameter aa breaks scale invariance, and the two zero modes are now protected by the Atiyah–Patodi–Singer (APS) index [26], which equals two for every aa (End Matter).

The gaugino condensate. What EH computes.

With fixed data at infinity, the EH path integral prepares a state ∣ΨEH⟩∈H(RP3)\lvert\Psi_{\mathrm{EH}}\rangle\in\mathcal{H}(\mathbb{RP}^{3}) on the asymptotic RP3\mathbb{RP}^{3}. Inserting O=tr⁡(λλ)≡tr⁡(ϵαβλβλα)\mathcal{O}=\operatorname{tr}(\lambda\lambda)\equiv\operatorname{tr}(\epsilon^{\alpha\beta}\lambda_{\beta}\lambda_{\alpha}) gives ∣ΨEH[O]⟩\lvert\Psi_{\mathrm{EH}}[\mathcal{O}]\rangle. A semiclassical computation fixes the asymptotic holonomy, so it evaluates the sector components Z(ℓ)[⋅]=⟨ℓ∣ΨEH[⋅]⟩Z^{(\ell)}[\cdot]=\langle\ell\vert\Psi_{\mathrm{EH}}[\cdot]\rangle. Since the bolt carries CS=0\mathrm{CS}=0, a saddle of charge QQ ends in the sector with CS≡Q\mathrm{CS}\equiv Q: saddles with Q∈ZQ\in\mathbb{Z} end in ℓ=0\ell=0, and saddles with Q∈12+ZQ\in\frac{1}{2}+\mathbb{Z} end in ℓ=1\ell=1. A saddle of charge QQ carries 4Q4Q gaugino zero modes. The trivial saddle has none and contributes to Z(0)[1]Z^{(0)}[1]. The Q=12Q=\frac{1}{2} saddle has two, saturated by O\mathcal{O}, and contributes to Z(1)[O]Z^{(1)}[\mathcal{O}]. XX-charge conservation makes this exact,

Z(0)[O]=0=Z(1)[1],Z(1)[O]≠0≠Z(0)[1],(17)Z^{(0)}[\mathcal{O}]=0=Z^{(1)}[1],\qquad Z^{(1)}[\mathcal{O}]\ne0\ne Z^{(0)}[1], \tag*{(17)}

which is the EH counterpart of O\mathcal{O} being off-diagonal in ℓ\ell, Eq. (14).

From sectors to vacua.

The condensate is therefore not defined sector by sector, ⟨ℓ∣O∣ℓ⟩=0\langle\ell\vert\mathcal{O}\vert\ell\rangle=0; it needs a bra that overlaps both sectors, namely a vacuum. The Witten index of SU(2)SU(2) SYM theory is two [27], and by Eq. (5) XX and UU cannot be diagonalized simultaneously on the two vacua. In the confining phase, the clustering vacua are the UU eigenstates, U∣k⟩=(−1)k∣k⟩U\lvert k\rangle=(-1)^k\lvert k\rangle (the standard picture, not derived here). Since ∣1⟩=U∣0⟩\lvert1\rangle=U\lvert0\rangle, their components obey ⟨k∣1⟩=(−1)k⟨k∣0⟩\langle k\vert1\rangle=(-1)^k\langle k\vert0\rangle, and

⟨O⟩k=⟨k∣ΨEH[O]⟩⟨k∣ΨEH[1]⟩=(−1)kZ(1)[O]Z(0)[1].(18)\langle\mathcal{O}\rangle_k=\frac{\langle k\vert\Psi_{\mathrm{EH}}[\mathcal{O}]\rangle}{\langle k\vert\Psi_{\mathrm{EH}}[1]\rangle}=(-1)^k\frac{Z^{(1)}[\mathcal{O}]}{Z^{(0)}[1]}. \tag*{(18)}

The anomaly fixes the relative sign: since XUX−1=−UXUX^{-1}=-U, XX maps ∣k⟩\lvert k\rangle to ∣k+1⟩\lvert k+1\rangle, and since XOX−1=−OX\mathcal{O}X^{-1}=-\mathcal{O}, ⟨O⟩k+1=−⟨O⟩k\langle\mathcal{O}\rangle_{k+1}=-\langle\mathcal{O}\rangle_k. The magnitude is the ratio of the Q=12Q=\frac{1}{2} saddle, with its zero modes saturated by O\mathcal{O}, to the trivial saddle. It is the EH counterpart of the constant CC, now computed on a supersymmetric background.

From EH to flat space.

Two Ward identities make Eq. (18) a flat-space vacuum quantity. First, ∂μO\partial_\mu\mathcal{O} is Q‾\overline{Q}-exact, so O\mathcal{O} can be inserted far from the bolt, where the geometry is flat. Second, the metric derivative ∂agμν\partial_a g_{\mu\nu} is a normalizable transverse traceless mode whose coupling to the stress tensor TμνT_{\mu\nu} is Q‾\overline{Q}-exact, so the result does not depend on the bolt size aa [14] (End Matter). The EH condensate was first studied in Ref. [28]. Its full collective-coordinate evaluation with Pauli–Villars regularization [14] gives

⟨tr⁡(λλ)⟩k=16π2Λk3,Λk3≡μ3g2e−4π2g2+i2(θ+2πk).(19)\langle\operatorname{tr}(\lambda\lambda)\rangle_k=16\pi^2\Lambda_k^3,\qquad\Lambda_k^3\equiv\frac{\mu^3}{g^2}e^{-\frac{4\pi^2}{g^2}+\frac{i}{2}(\theta+2\pi k)}. \tag*{(19)}

This is precisely the weak-coupling instanton (WCI) value. On R4\mathbb{R}^4 the BPST instanton has four zero modes and cannot saturate O\mathcal{O}, so the strong-coupling instanton (SCI) route extracts ⟨O⟩\langle\mathcal{O}\rangle from ⟨OO⟩\langle\mathcal{O}\mathcal{O}\rangle by clustering, and obtains a value smaller by 4/5\sqrt{4/5} [29, 30, 31]. The single EH saddle needs no clustering and sides with WCI, consistent with Refs. [31, 32, 33, 34, 35, 36].

Generalizations.

The construction extends to SU(N)SU(N) on lens space L(p,1)≡S3/ZpL(p,1)\equiv S^3/\mathbb{Z}_p, where Zp\mathbb{Z}_p acts on C2≅R4\mathbb{C}^2\cong\mathbb{R}^4 by (z1,z2)↦ζ(z1,z2)(z_1,z_2)\mapsto\zeta(z_1,z_2), ζ=e2πi/p\zeta=e^{2\pi i/p}, i.e. ψ↦ψ+4π/p\psi\mapsto\psi+4\pi/p. A center transformation along cc must obey zp=1z^p=1, so only the subgroup Zd\mathbb{Z}_d acts, with d=gcd⁡(p,N)d=\gcd(p,N). The flat sectors are hc=diag⁡(ζm1,…,ζmN)h_c=\operatorname{diag}(\zeta^{m_1},\ldots,\zeta^{m_N}), with CS≡−∑imi2/2p\mathrm{CS}\equiv-\sum_i m_i^2/2p for a lift with ∑imi=0\sum_i m_i=0 [37, 38, 39]. The generator of Zd\mathbb{Z}_d shifts every mim_i by p/dp/d. After re-lifting, this shifts CS uniformly by

ΔCS=−pN(N+1)2d2(mod1).(20)\Delta\mathrm{CS}=-\frac{pN(N+1)}{2d^2}\pmod{1}. \tag*{(20)}

Since d∣pd\mid p and d∣Nd\mid N, ΔCS∈12Z\Delta\mathrm{CS}\in\frac{1}{2}\mathbb{Z}. A purely temporal background, B=a∪dtB=a\cup dt, has ∫B∪B=0(modN)\int B\cup B=0\pmod N; only the Pontryagin-square refinement survives, through the torsion of L(p,1)L(p,1), so it detects at most a Z2\mathbb{Z}_2 part of the mixed anomaly. The phase XUX−1U−1=U′U−1U′−1=e2πiΔCSXUX^{-1}U^{-1}=U'U^{-1}U'^{-1}=e^{2\pi i\Delta\mathrm{CS}} is −1-1 if and only if

ν2(p)=ν2(N)≥1,(21)\nu_2(p)=\nu_2(N)\ge1, \tag*{(21)}

where ν2(n)\nu_2(n) is the exponent of the largest power of 22 dividing nn. When Eq. (21) does not hold, the lens space is anomaly blind. On RP3\mathbb{R}P^3 the criterion gives N≡2(mod4)N\equiv2\pmod4.

Torsion need not come from curved space. For SU(2)SU(2), the anomaly is visible on a closed three-manifold MM whenever some a∈H1(M;Z2)a\in H^1(M;\mathbb{Z}_2) has ∫Ma3≠0\int_M a^3\ne0 [40]. Among flat manifolds [41], we find that this holds for the quarter-turn space, a cubic lattice that is periodic in xx and yy and closed in zz after a 90∘90^\circ rotation, but not for T3\mathbb{T}^3, the half-turn space, or the Hantzsche–Wendt space. We therefore expect that Hamiltonian lattice gauge theory, or a quantum simulator, with this screw boundary condition would exhibit the exact twofold degeneracies of SU(2)SU(2) SYM theory and of YM theory at θ=π\theta=\pi without any ’t Hooft flux, with the other flat manifolds as built-in controls.

Acknowledgments

We would like to thank İñaki García Etxebarria, Erich Poppitz, and Yuya Tanizaki for discussions and valuable comments on the manuscript. Also, we would like to thank Erich Poppitz for the hospitality at the University of Toronto, where part of this project was conducted. Opus 5.5 Claude by Anthropic and Astra-6 ChatGPT by OpenAI were used for calculation checks and edits. This work is supported by the STFC consolidated grant UKRI14226.

End Matter

Conventions and the Chern–Simons invariant on RP3\mathbb{R}\mathbb{P}^{3}

We follow Ref. [14]: Hermitian gauge fields with F=dA+iA∧AF=dA+iA\wedge A, color Pauli matrices τa\tau_a, spin matrices σ^a=(I2,iσ⃗)\hat{\sigma}^{a}=(I_2,i\vec{\sigma}), and topological charge

Q=18π2∫tr⁡F∧F.Q=\frac{1}{8\pi^2}\int\operatorname{tr}F\wedge F.

The left-invariant one-forms on S3S^3 are

σx=12(sin⁡ψ dθ−sin⁡θcos⁡ψ dϕ),\sigma_x=\frac{1}{2}\left(\sin\psi\,d\theta-\sin\theta\cos\psi\,d\phi\right),
σy=12(−cos⁡ψ dθ−sin⁡θsin⁡ψ dϕ),\sigma_y=\frac{1}{2}\left(-\cos\psi\,d\theta-\sin\theta\sin\psi\,d\phi\right),
σz=12(dψ+cos⁡θ dϕ),(22)\sigma_z=\frac{1}{2}\left(d\psi+\cos\theta\,d\phi\right), \tag*{(22)}

with dσx=2σy∧σzd\sigma_x=2\sigma_y\wedge\sigma_z (cyclic) and σx∧σy∧σz=−18sin⁡θ dθ∧dϕ∧dψ\sigma_x\wedge\sigma_y\wedge\sigma_z=-\frac{1}{8}\sin\theta\,d\theta\wedge d\phi\wedge d\psi, so that ∫RP3σx∧σy∧σz=−π2\int_{\mathbb{R}\mathbb{P}^{3}}\sigma_x\wedge\sigma_y\wedge\sigma_z=-\pi^2.

We parametrize

g=eiϕτ3/2eiθτ2/2ei(ψ+π)τ3/2,(23)g=e^{i\phi\tau_3/2}e^{i\theta\tau_2/2}e^{i(\psi+\pi)\tau_3/2}, \tag*{(23)}

with 0≤θ≤π0\leq\theta\leq\pi, 0≤ϕ<2π0\leq\phi<2\pi, 0≤ψ<4π0\leq\psi<4\pi; a direct computation gives g−1dg=iτaσag^{-1}dg=i\tau_a\sigma_a, the shift ψ→ψ+π\psi\to\psi+\pi aligning the Maurer–Cartan form with Eq. (22). With CS⁡(A)=18π2∫tr⁡(A∧dA+2i3A∧A∧A)\operatorname{CS}(A)=\frac{1}{8\pi^2}\int\operatorname{tr}\left(A\wedge dA+\frac{2i}{3}A\wedge A\wedge A\right), so that dCS⁡=18π2tr⁡F∧Fd\operatorname{CS}=\frac{1}{8\pi^2}\operatorname{tr}F\wedge F, a pure gauge −ih−1dh-ih^{-1}dh has CS⁡=124π2∫tr⁡(h−1dh)3\operatorname{CS}=\frac{1}{24\pi^2}\int\operatorname{tr}(h^{-1}dh)^3. Using tr⁡(τaτbτc)=2iϵabc\operatorname{tr}(\tau_a\tau_b\tau_c)=2i\epsilon_{abc}, tr⁡(g−1dg)3=12σx∧σy∧σz\operatorname{tr}(g^{-1}dg)^3=12\sigma_x\wedge\sigma_y\wedge\sigma_z. Since A1=−igdg−1A_1=-igdg^{-1} corresponds to h=g−1h=g^{-1},

CS⁡(A1)=−1224π2∫σx∧σy∧σz={1,S3: ψ∈[0,4π),12,RP3: ψ∈[0,2π).(24)\operatorname{CS}(A_1)=-\frac{12}{24\pi^2}\int\sigma_x\wedge\sigma_y\wedge\sigma_z= \begin{cases} 1, & S^3:\ \psi\in[0,4\pi),\\ \frac{1}{2}, & \mathbb{R}\mathbb{P}^{3}:\ \psi\in[0,2\pi). \end{cases} \tag*{(24)}

Center operator and the vanishing traces

Because g(ψ+2π)=−g(ψ)g(\psi+2\pi)=-g(\psi) and Ad⁡(−g)=Ad⁡(g)\operatorname{Ad}(-g)=\operatorname{Ad}(g), UU maps antipodally invariant adjoint configurations to antipodally invariant ones. The holonomy transforms as h↦−ghg−1h\mapsto-ghg^{-1}, and U(A0)=A1U(A_0)=A_1. On a closed three-manifold CS⁡(AU)=CS⁡(A)+CS⁡(A1)\operatorname{CS}(A^U)=\operatorname{CS}(A)+\operatorname{CS}(A_1), which gives Eq. (4). Then UXU−1=Xeiπ(−4)12=−XUXU^{-1}=Xe^{i\pi(-4)\frac{1}{2}}=-X and UT′U−1=e2πi(CS⁡+12)T=−T′UT'U^{-1}=e^{2\pi i(\operatorname{CS}+\frac{1}{2})}T=-T'.

For the traces, note that U2=1U^2=1 and [U,H]=0[U,H]=0, so in a basis that diagonalizes both, Tr⁡[Ue−βH]=∑n±e−βEn\operatorname{Tr}[Ue^{-\beta H}]=\sum_n\pm e^{-\beta E_n} is real. In SYM theory, XX is unitary and commutes with HH, so Tr⁡[Ue−βH]=Tr⁡[XUX−1e−βH]=−Tr⁡[Ue−βH]=0\operatorname{Tr}[Ue^{-\beta H}]=\operatorname{Tr}[XUX^{-1}e^{-\beta H}]=-\operatorname{Tr}[Ue^{-\beta H}]=0. In YM theory at θ=π\theta=\pi, T′T' is antiunitary and commutes with HH, and T′UT′−1=−UT'UT'^{-1}=-U. For an antiunitary operator, Tr⁡[T′AT′−1]=Tr⁡[A]∗\operatorname{Tr}[T'AT'^{-1}]=\operatorname{Tr}[A]^*, so Tr⁡[Ue−βH]∗=−Tr⁡[Ue−βH]\operatorname{Tr}[Ue^{-\beta H}]^*=-\operatorname{Tr}[Ue^{-\beta H}]. The trace is both real and imaginary, hence zero.

The tunneling solution

The instanton (9) survives the quotient because of its symmetries. Rotations act as X→LXR†X\to LXR^\dagger, (L,R)∈SU(2)L×SU(2)R(L,R)\in SU(2)_L\times SU(2)_R, i.e., g→LgR†g\to LgR^\dagger. Eq. (9) is exactly SU(2)RSU(2)_R invariant, while an SU(2)LSU(2)_L rotation equals the constant gauge rotation A→LAL−1A\to LAL^{-1}, the familiar locking of the undotted spinor index to color. The element L=−I2L=-I_2, with R=I2R=I_2, is the reflection x→−xx\to-x, and its gauge action is central and therefore trivial. Hence Eq. (9) is invariant under x→−xx\to-x with no gauge transformation. The same fact makes the orientation moduli SU(2)/Z2≅SO(3)≅RP3SU(2)/\mathbb{Z}_2\cong SO(3)\cong\mathbb{R}\mathbb{P}^{3}, so the four centered moduli, the size and three SO(3)SO(3) orientations, form the cone R4/Z2\mathbb{R}^{4}/\mathbb{Z}_2 itself.

Since A1=−igdg−1A_1=-igdg^{-1} is flat, dA1=−iA1∧A1dA_1=-iA_1\wedge A_1. For A=f(τ)A1A=f(\tau)A_1 this gives

F=f′ dτ∧A1+i(f2−f)A1∧A1.(25)F=f'\,d\tau\wedge A_1+i(f^2-f)A_1\wedge A_1. \tag*{(25)}

Write A1=τaθaA_1=\tau_a\theta^a. Flatness implies dθa=ϵabcθb∧θcd\theta^a=\epsilon_{abc}\theta^b\wedge\theta^c, so the θa\theta^a are right-invariant forms with the normalization of Eq. (22), and the round metric on SR3S_R^3 is R2∑a(θa)2R^2\sum_a(\theta^a)^2. Using τbτc=δbc+iϵbcaτa\tau_b\tau_c=\delta_{bc}+i\epsilon_{bca}\tau_a,

F=τa[f′RE0∧Ea+2f(1−f)R2Eb∧Ec],(26)F=\tau_a\left[\frac{f'}{R}E^0\wedge E^a+\frac{2f(1-f)}{R^2}E^b\wedge E^c\right], \tag*{(26)}

with E0=dτE^0=d\tau, Ea=RθaE^a=R\theta^a, and (a,b,c)(a,b,c) cyclic. The right-invariant forms obey θb=−(Ad⁡g)baσa\theta^b=-(\operatorname{Ad}g)_{ba}\sigma^a, hence θ1∧θ2∧θ3=−σx∧σy∧σz\theta^1\wedge\theta^2\wedge\theta^3=-\sigma_x\wedge\sigma_y\wedge\sigma_z. Since ∣x∣=ρeτ/R|x|=\rho e^{\tau/R}, the orientation of Ref. [14], vol⁡∝−dr∧σx∧σy∧σz\operatorname{vol}\propto-dr\wedge\sigma_x\wedge\sigma_y\wedge\sigma_z, is vol⁡∝+E0∧E1∧E2∧E3\operatorname{vol}\propto+E^0\wedge E^1\wedge E^2\wedge E^3, so ∗(E0∧Ea)=Eb∧Ec*(E^0\wedge E^a)=E^b\wedge E^c. Self-duality is then f′=2f(1−f)/Rf'=2f(1-f)/R, solved by f=[1+e−2(τ−τ0)/R]−1f=[1+e^{-2(\tau-\tau_0)/R}]^{-1}. The opposite sign gives the anti-self-dual event from A1A_1 to A0A_0. Equivalently, the configuration has conformal image of Eq. (9) with τ0=Rln⁡(ρ/R)\tau_0=R\ln(\rho/R), since a Weyl rescaling preserves the Hodge star on two-forms. Its charge is Q=CS⁡(A1)−CS⁡(A0)=12Q=\operatorname{CS}(A_1)-\operatorname{CS}(A_0)=\frac{1}{2}. We use right-invariant forms because they make the regular-gauge instanton (9) self-dual in this orientation; that instanton is smooth at the orbifold point and asymptotes to A1A_1. The left-invariant representative −ig−1dg=τaσa-ig^{-1}dg=\tau_a\sigma_a of the same sector has CS⁡=−12\operatorname{CS}=-\frac{1}{2} and is related to A1A_1 by a degree-one large gauge transformation. In it, the self-dual event is the singular-gauge solution ρ2/(x2+ρ2)τaσa\rho^2/(x^2+\rho^2)\tau_a\sigma_a from ℓ=1\ell=1 to ℓ=0\ell=0, the type to which the EH solution (16) belongs. The two descriptions are related by A→g−1Ag−ig−1dgA\to g^{-1}Ag-ig^{-1}dg, which maps fτaθaf\tau_a\theta^a to (1−f)τaσa(1-f)\tau_a\sigma_a. On S3S^3 this is a large gauge transformation; on RP3\mathbb{R}\mathbb{P}^3 it is the center map UU (up to the allowed transformation g−2g^{-2}), so the two events are related by symmetry and carry equal weight.

Zero modes

For the adjoint representation of SU(N)SU(N) on EH with asymptotic holonomy hh, the APS index is

Iadj=2NQ+−(N2−1)+sgn⁡(Hol⁡)8,(27)\mathcal{I}_{\mathrm{adj}}=2NQ+\frac{-(N^2-1)+\operatorname{sgn}(\operatorname{Hol})}{8}, \tag*{(27)}

where the second term combines the gravitational and η\eta-invariant contributions. Here sgn⁡(Hol⁡)=∑weights(±1)\operatorname{sgn}(\operatorname{Hol})=\sum_{\text{weights}}(\pm1) counts the adjoint components that are periodic (+1)(+1) or antiperiodic (−1)(-1) around cc. For SU(2)SU(2) with central h=−Ih=-I, all three components are periodic, so sgn⁡(Hol⁡)=3\operatorname{sgn}(\operatorname{Hol})=3 and Iadj=4⋅12+1=3\mathcal{I}_{\mathrm{adj}}=4\cdot\frac{1}{2}+1=3, independent of aa. Also the explicit EH modes [14] have the profile rt2/(r2+t2)2rt^2/(r^2+t^2)^2 of the superconformal modes, not the profile ρ2/(r2+ρ2)2\rho^2/(r^2+\rho^2)^2 of the supersymmetric ones. On R4\mathbb{R}^4 the BPST field strength scales as ∣F∣∝ρ2/(r2+ρ2)2|F| \propto\rho^2/(r^2+\rho^2)^2. The supersymmetric modes λ=Fξ\lambda= F\xi, with ξ\xi constant, inherit this profile and are nonzero at the origin. The superconformal modes λ=Fϵ(x)\lambda= F\epsilon(x), with ϵ(x)=xaσaϵˉ\epsilon(x) = x_a\sigma^a\bar{\epsilon} linear in xx, carry an extra factor of rr, ∣λ∣∝rρ2/(r2+ρ2)2|\lambda| \propto r\rho^2/(r^2+\rho^2)^2, and vanish at the origin; this factor of xx is also what makes them odd under x→−xx \to-x. With the size tt in place of ρ\rho, the EH modes at a→0a \to0 reproduce the second profile [14], confirming that the two surviving zero modes are the superconformal ones.

independence of aa

In the coordinate ϱ4=r4−a4\varrho^4=r^4-a^4, the EH metric is ds2=ϱ2R2dϱ2+R2(σx2+σy2)+ϱ4R2σz2ds^2=\frac{\varrho^2}{R^2}d\varrho^2+R^2(\sigma_x^2+\sigma_y^2)+\frac{\varrho^4}{R^2}\sigma_z^2 with R4=ϱ4+a4R^4=\varrho^4+a^4. At fixed ϱ\varrho, ∂aR=a3/R3\partial_aR=a^3/R^3, and hμν=∂agμνh_{\mu\nu}=\partial_ag_{\mu\nu} is, in the orthonormal frame,

hab=2a3R4diag⁡(−1,1,1,−1).(28)h_{ab}=\frac{2a^3}{R^4}\operatorname{diag}(-1,1,1,-1). \tag*{(28)}

We checked that hh is traceless and divergence-free, and that ∫g habhab<∞\int\sqrt{g}\,h_{ab}h^{ab}<\infty. For a symmetric transverse traceless hh, the Ferrara–Zumino multiplet gives

12hμνTμν={Qˉ,V}+∇ν(⋯ )ν.(29)\frac{1}{2}h^{\mu\nu}T_{\mu\nu}=\{\bar{Q},V\}+\nabla_\nu(\cdots)^\nu. \tag*{(29)}

The surface term falls off as a3/r4a^3/r^4 against an area ∝r3\propto r^3, and the bra, the ket, and OO are Qˉ\bar{Q}-closed. Hence ∂a⟨O⟩=0\partial_a\langle O\rangle=0; the trace anomaly drops out because hh is traceless. The details on the derivation of Ward identities are in Ref. [14].

References

  1. [1]D. Gaiotto, A. Kapustin, N. Seiberg and B. Willett, JHEP 02, 172 (2015), arXiv:1412.5148.DOI
  2. [2]D. Gaiotto, A. Kapustin, Z. Komargodski and N. Seiberg, JHEP 05, 091 (2017), arXiv:1703.00501.DOI
  3. [3]H. Shimizu and K. Yonekura, Phys. Rev. D 97, no.10, 105011 (2018) doi:10.1103/PhysRevD.97.105011 [arXiv:1706.06104 [hep-th]].DOI
  4. [4]Z. Komargodski, T. Sulejmanpasic and M. Ünsal, Phys. Rev. B 97, no.5, 054418 (2018) doi:10.1103/PhysRevB.97.054418 [cond-mat.str-el].DOI
  5. [5]G. ’t Hooft, Nucl. Phys. B 153, 141 (1979).
  6. [6]P. van Baal, Commun. Math. Phys. 85, 529 (1982) doi:10.1007/BF01403503DOI
  7. [7]P. van Baal, PhD thesis, 1984.
  8. [8]A. Gonzalez-Arroyo, [arXiv:hep-th/9807108 [hep-th]].
  9. [9]P. van Baal, doi:10.1142/9789812810458_0021 [arXiv:hep-ph/0008206 [hep-ph]].
  10. [10]A. A. Cox, E. Poppitz and F. D. Wandler, JHEP 10, 069 (2021) doi:10.1007/JHEP10(2021)069 [arXiv:2106.11442 [hep-th]].DOI
  11. [11]M. M. Anber and S. Y. L. Chan, JHEP 03, 169 (2024) doi:10.1007/JHEP03(2024)169 [arXiv:2311.07662 [hep-th]].DOI
  12. [12]Y. Hayashi and Y. Tanizaki, PTEP 2026, no.8, 083B04 (2026) doi:10.1093/ptep/ptag134 [arXiv:2606.17708 [hep-th]].DOI
  13. [13]T. Egucchi and A. J. Hanson, Phys. Lett. B 74, 249 (1978).
  14. [14]M. M. Anber, [arXiv:2610.00486 [hep-th]].arxiv.org/abs/2610.00486
  15. [15]D. S. Freed, Adv. Math. 113, 237-303 (1995) doi:10.1006/aima.1995.1039 [arXiv:hep-th/9206021 [hep-th]].DOI
  16. [16]D. S. Freed, G. W. Moore and G. Segal, Commun. Math. Phys. 271, 247-274 (2007) doi:10.1007/s00220-006-0181-3 [arXiv:hep-th/0605198 [hep-th]].DOI
  17. [17]D. S. Freed, G. W. Moore and G. Segal, Annals Phys. 322, 236-285 (2007) doi:10.1016/j.aop.2006.07.014 [arXiv:hep-th/0605200 [hep-th]].DOI
  18. [18]F. B. Christensen, I. García Etxebarria and E. Leung, JHEP 02, 177 (2026) doi:10.1007/JHEP02(2026)177 [arXiv:2510.19935 [hep-th]].DOI
  19. [19]M. Garcia Perez and A. Gonzalez-Arroyo, J. Phys. A 26, 2667-2678 (1993) doi:10.1088/0305-4470/26/11/015 [arXiv:hep-lat/9206016 [hep-lat]].DOI
  20. [20]J. Dasilva, Galan and M. García Pérez, JHEP 12, 109 (2022) doi:10.1007/JHEP12(2022)109 [arXiv:2208.07133 [hep-th]].
  21. [21]M. Unsal, Phys. Rev. D 86, 105012 (2012) doi:10.1103/PhysRevD.86.105012 [arXiv:1201.6426 [hep-th]].DOI
  22. [22]M. M. Anber, JHEP 10, 183 (2026) [arXiv:2605.26224 [hep-th]].arxiv.org/abs/2605.26224
  23. [23]P. B. Kronheimer and H. Nakajima, Math. Ann. 288, 263 (1990).
  24. [24]M. Bianchi, F. Fucito, G. C. Rossi and M. Martinelli, Phys. Lett. B 359, 49 (1995).
  25. [25]M. Bianchi, F. Fucito, G. C. Rossi and M. Martinelli, Nucl. Phys. B 473, 367 (1996), arXiv:hep-th/9601162.DOI
  26. [26]M. F. Atiyah, V. K. Patodi and I. M. Singer, Math. Proc.
  27. [27]Cambridge Phil. Soc. 77, 43 (1975). [27] E. Witten, Nucl. Phys. B 202, 253 (1982).
  28. [28]M. Bianchi, F. Fucito and G. C. Rossi, Phys. Lett. B 359, 56 (1995), arXiv:hep-th/9507003.DOI
  29. [29]D. Amati, K. Konishi, Y. Meurice, G. C. Rossi and G. Veneziano, Phys. Rept. 162, 169 (1988).
  30. [30]D. Finnell and P. Pouliot, Nucl. Phys. B 453, 225 (1995), arXiv:hep-th/9503115.DOI
  31. [31]T. J. Hollowood, V. V. Khoze, W. Lee and M. P. Mattis, Nucl. Phys. B 570, 241 (2000), arXiv:hep-th/9904116.
  32. [32]N. M. Davies, T. J. Hollowood, V. V. Khoze and M. P. Mattis, Nucl. Phys. B 559, 123 (1999), arXiv:hep-th/9905015.DOI
  33. [33]M. M. Anber and E. Poppitz, JHEP 01, 118 (2023) doi:10.1007/JHEP01(2023)118 [arXiv:2210.13568 [hep-th]].DOI
  34. [34]M. M. Anber and E. Poppitz, JHEP 02, 114 (2025) doi:10.1007/JHEP02(2025)114 [arXiv:2408.16058 [hep-th]].
  35. [35]C. Bonanno, P. Butti, M. García Pérez, A. González-Arroyo, K. I. Ishikawa and M. Okawa, Phys. Rev. D 110, no.7, 074507 (2024) doi:10.1103/PhysRevD.110.074507 [arXiv:2406.08995 [hep-th]].DOI
  36. [36]Y. Hayashi and Y. Tanizaki, [arXiv:2611.00639 [hep-th]].arxiv.org/abs/2611.00639
  37. [37]P. Kirk and E. Klassen, Math. Ann. 287, 343 (1990).
  38. [38]S. S. Razamat and B. Willett, Commun. Math. Phys. 334, no.2, 661-696 (2015) doi:10.1007/s00220-014-2111-0 [arXiv:1307.4381 [hep-th]].DOI
  39. [39]Z. Duan, Q. Jia and S. Lee, JHEP 01, 161 (2025) doi:10.1007/JHEP01(2025)161 [arXiv:2410.10036 [hep-th]].DOI
  40. [40]V. G. Turaev, Math. USSR-Sb. 48, 65 (1984).
  41. [41]J. H. Conway and J. P. Rossetti, [arXiv:math/0311476 [math.DG]].

Paper details

vlnc:W_5ef36dsg
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Astra-6 ChatGPT

View statement
Looking for related papers…
Contents