Reading Pestun's four-sphere localization paper

The Gaussian answer is the endpoint of three separate calculations: the physical action on the localization locus, the determinant of ordinary fluctuations, and the contributions allowed at the two poles. Reading only the final matrix integral hides the distinction between them.

This guide follows Vasily Pestun, Localization of gauge theory on a four-sphere and supersymmetric Wilson loops, arXiv:0712.2824v3, 20 September 2012, published in Communications in Mathematical Physics 313 (2012), 71–129. The manuscript’s opening date says December 2007; use the v3 header to identify the version being read. Its footnote 9 on page 39 explicitly records a mass-parameter correction introduced in v2. The corrected distinction is essential below.

Start with supersymmetric Wilson lines and circles, N=2 vectors and full hypermultiplets, and the general localization argument. Massless four-dimensional N=4 SYM at its conformal origin has sixteen real Poincaré and sixteen conformal supercharges. Organizing it as an N=2 vector plus an adjoint hypermultiplet, and choosing one odd localization charge, does not change that full theory. Giving the hyper a physical mass produces the N=2-star deformation, whose flat parent has eight Poincaré charges; it is a different theory, not a different notation for massless N=4. The source constructs an eight-generator rigid sphere subalgebra compatible with that mass, then selects one charge. Euclidean spinor labels are complexified, and the seven auxiliaries close this selected operation rather than every supercharge.

Read the argument in this order

First read the introduction’s (1.1)–(1.4) to identify the observable and the three factors. Return to them after each row below. In these page references the printed page and one-based PDF page agree.

The following passage uses the complete sequence from the sphere background through the planar limit. The linked sections supply the canonical derivations; the companion tells you what to extract from the original paper.

Source passage	Question to answer	Canonical calculation
(1.1), p. 2; (2.18)–(2.19), p. 13	Which scalar accompanies the connection, and which spinor preserves the equatorial loop?	The circle's supersymmetry condition
(2.2)–(2.6), p. 9; (2.12)–(2.14), pp. 11–12; Appendix B, pp. 44–45	How do the conformal spinor and curvature term enter?	Solve the sphere spinor equations
(2.26)–(2.33), pp. 15–16; Appendix C, pp. 45–50	What closes off shell for the selected charge, and what does its square do?	Identify the selected square and auxiliaries
(3.16)–(3.18), pp. 19–20	Why is a tempting extra smooth scalar branch excluded? Which auxiliary survives?	The Bianchi-identity test
(3.19), p. 21	Why is the classical coefficient proportional to three, rather than two?	Evaluate the physical auxiliary action
(4.20)–(4.22), p. 27; (4.39)–(4.46), pp. 30–32	Which unpaired modes determine the ratio, and why are the two pole expansions different?	Normal fluctuations and their signed index
(4.47)–(4.50), p. 33; (4.58)–(4.59), pp. 34–35	When do vector and hypermultiplet weights actually cancel?	The determinant product
Section 5, pp. 36–39; especially (5.11) and footnote 9	Which mass vanishes in N=4, and what happens to the polar sectors?	The shifted mass and the nonempty sectors
Return to (1.2)–(1.4), p. 3	How are the trace and the finite-dimensional integral normalized?	Finite-color evaluation and central-mode removal

Appendix A, pp. 40–44, fixes the Clifford blocks used in the source; it is a lookup tool while reading the selected spinor. Appendix B solves the conformal spinor equations. Appendix C supplies the auxiliary closure. Appendix D, pp. 50–54, explains the index machinery behind Section 4.4. You can first use the explicitly stated index result and then return to Appendix D to study its proof. An equivariant index counts unpaired modes with signs and rotation weights; it is not an ordinary total count of bosonic and fermionic fields. A transversally elliptic operator is elliptic for nonzero covectors perpendicular to the rotation orbits. This is why the rotation and its fixed points remain part of the calculation.

The positive- and negative-power expansions in (4.39)–(4.44) are prescribed distributions. Treating them as the same unexpanded rational function loses information at the identity rotation. Conversely, cancellation of the adjoint vector and hypermultiplet contributions at each nonzero weight really does make the normal determinant trivial. It leaves the finite gauge-orbit Jacobian in place; see the residual Weyl measure.

Translate before substituting

Our principal target is the fundamental loop of $SU(N)$, $N\geq2$, with rank $N-1$, $\operatorname{Tr}_N(t^at^b)=\delta^{ab}/2$ and $\lambda=g^2N$. The source also considers general representations and explicitly uses $U(N)$ in its instanton sum. Those are useful extensions, not interchangeable choices of global theory.

Source notation or choice	Translation used in the lessons
Sphere radius $r$	$\ell$; the round local frequencies are $\epsilon_1=\epsilon_2=1/\ell$
Unnormalized $\operatorname{tr}_R$ in (1.1)	Divide by $\dim R$ for the normalized loop; divide by $N$ in the fundamental
Compact anti-Hermitian algebra with positive pairing $-\operatorname{Tr}$	Hermitian absorbed fields, with ordinary positive $\operatorname{Tr}_N a^2$ on the scalar locus
Continued scalar $\Phi_0^{P,E}$ on the locus	$\Phi_0^{P,E}=-ia$, so $i\Phi_0^{P,E}=a$ and the loop contains $e^{2\pi\ell a}$
Real eigenvalues $a_E$ in (4.47)	Eigenvalues of the Hermitian $a$; the analytic pole input is $\mathfrak a_i=ia_i$
$H_P(z)=G(1+z)G(1-z)$ in (4.49)	$H(x)=H_P(ix)=G(1+ix)G(1-ix)$, with $x=\ell\alpha(a)$
Analytic mass $m$ and shifted mass $\widetilde m$	$m_P=i m_E$ on the stated real-$m_E$ contour, and $m_{\rm eq}=\widetilde m=m_P+(\epsilon_1+\epsilon_2)/2$
Source instanton charge and theta angle	With the fixed orientation and $F_P=-iF_H$, $k_P=-k_{\rm site}$ and $\theta_P=-\theta_{\rm site}$; retain north/south labels when assigning $q$ and its conjugate

The scalar entry is an on-locus translation, not permission to erase the factors of $i$ in the full transformation laws. The four-sphere convention dictionary supplies those laws and the integration contour. In the source’s Section 5 and footnote 7, the literal matrix trace defining $k_P$ differs from its positive Lie-algebra pairing. This accounts for the charge sign in the table. The mass relation is also a Euclidean source dictionary, not a universal identification of a positive Lorentzian particle mass with a complex instanton parameter.

For orientation, the general result has the structure

$$\langle W_R\rangle= \frac{\displaystyle\int_{\mathfrak t} da\, \Delta(a)^2 e^{-S_{\rm cl}(a)} Z_{\rm normal}(a,m_P)Z_N Z_S\,w_R(a)} {\displaystyle\int_{\mathfrak t} da\, \Delta(a)^2 e^{-S_{\rm cl}(a)} Z_{\rm normal}(a,m_P)Z_N Z_S}, \tag{PC12.1}$$

where $\mathfrak t$ is the Hermitian Cartan subalgebra, $\Delta(a)=\prod_{\alpha>0}\alpha(a)$, $S_{\rm cl}=8\pi^2\ell^2\operatorname{Tr}_N a^2/g^2$, and $w_R=\operatorname{Tr}_R e^{2\pi\ell a}/\dim R$. The same Weyl-group factor and measure constants cancel between numerator and denominator. $Z_N,Z_S$ are the two prescribed polar factors; on the physical contour they form the source’s absolute square. They are not inferred from the smooth locus alone.

For physical N=4, $m_P=0$ but $m_{\rm eq}=1/\ell$. The normal factor and polar factors then become one in the stated prescription. Setting $M=2\pi\ell a$ gives the Gaussian weight $e^{-2N\operatorname{Tr}M^2/\lambda}$ and the insertion $\operatorname{Tr}e^M/N$. The normalized answer is independent of $\ell$. An absolute partition function still needs its measure and background-counterterm convention; it should not be identified with a normalized loop.

Restore a two-box mass limit

The cancellation of all nonempty sectors is explained in the Gaussian-model lesson. Here we reconstruct a compressed step of Pestun (5.3)–(5.6) and (5.11): how a complete coefficient remains regular when individual fixed-point formulas have zero denominators.

Use the local $U(1)$ instanton-counting prescription as a small example. This is the resolved equivariant moduli-space calculation used by the source, not a claim that smooth ordinary Abelian instantons exist on $S^4$. Its fixed points are labeled by Young diagrams, arrangements of boxes with nonincreasing row lengths. Two boxes have a horizontal and a vertical arrangement. Write $\alpha=\epsilon_1$, $\beta=\epsilon_2$, $m=m_{\rm eq}$ in this section only, and define

$$F(E)=\frac{(E-m)(\alpha+\beta-E-m)}{E(\alpha+\beta-E)}. \tag{PC12.2}$$

Each pair of tangent weights $E,\alpha+\beta-E$ supplies this factor: the vector supplies the denominator and the adjoint hyper supplies the mass-shifted numerator. From (5.4), the two diagrams give

$$T_h=F(\beta)F(\beta-\alpha),\qquad T_v=F(\beta)F(2\beta). \tag{PC12.3}$$

The weights have a useful independent geometric check. Near the horizontal diagram, describe a length-two configuration in the complex plane by $y=u_0+u_1x$, $x^2=v_0+v_1x$. Its coordinate ring has basis $1,x$, so these four coefficients are independent local deformations. Under $x\mapsto t_1x$, $y\mapsto t_2y$, their weights are $\beta,\beta-\alpha,2\alpha,\alpha$. Exchanging $x,y$ gives the vertical chart with weights $\alpha,\alpha-\beta,2\beta,\beta$. They reproduce the two paired denominators in (PC12.3). No total field-count argument could supply these rotation weights.

Add both diagrams before imposing equal frequencies. Common denominators and cancellation give

$$\begin{aligned} Z_1&=\frac{(\alpha-m)(\beta-m)}{\alpha\beta},\\ Z_2=T_h+T_v &=\frac{(\alpha-m)(\beta-m) \left[4\alpha\beta-m(\alpha+\beta)+m^2\right]} {2\alpha^2\beta^2},\\ Z_{\rm pole}&=1+qZ_1+q^2Z_2+O(q^3). \end{aligned} \tag{PC12.4}$$

Neither complete coefficient has a pole at $\alpha=\beta\ne0$. For example, keep zero physical mass while regulating the frequencies by $\alpha=e+\delta$, $\beta=e-\delta$, $m=(\alpha+\beta)/2=e$, with $e>0$ and $|\delta|<e$. This is a regulator for the local formula, not a claim to have constructed a new unequal-frequency global sphere background. Then

$$Z_1=-\frac{\delta^2}{e^2-\delta^2},\qquad Z_2=\frac{\delta^2(4\delta^2-3e^2)} {2(e^2-\delta^2)^2}\ \longrightarrow\ 0. \tag{PC12.5}$$

Individually $T_h=-\delta/(4e)+O(\delta^2)$ and $T_v=+\delta/(4e)+O(\delta^2)$; their linear terms cancel. A literal substitution into the unresolved factors would instead encounter $0/0$. At the round point $e=1/\ell$, the continued coefficients verify the physical massless result through two boxes. The canonical argument establishes all sectors; this explicit calculation tests the limit operation rather than replacing that argument.

Contrast this with setting $m_{\rm eq}=0$. Equations (PC12.2)–(PC12.4) give unit weight for every diagram, $Z_1=1$ and $Z_2=2$. This is $m_P=-e$, away from the real-$m_E$ contour at finite radius, not physical N=4. Pestun (5.13) gives the corresponding partition-generating function. The correction footnote explains why v1’s identification of these two masses must not be used.

What the original argument establishes

Pestun’s introduction distinguishes the earlier Gaussian proposal from its localization derivation. A constant one-loop exchange on a circle and a sum of rainbow diagrams motivated the answer, but neither establishes the cancellation of every interaction or instanton sector. Sections 2–5 supply the chosen off-shell charge, positive deformation on a specified contour, smooth locus, gauge-fixed index and polar factors.

This is a physics localization argument with a regulator prescription, not a claim that every analytic issue of an infinite-dimensional integral has disappeared. In particular, the discussion on page 4 assumes a compatible small-instanton resolution on the sphere matching the local Nekrasov prescription near the poles. The contour and Ward-identity conditions are those of the general localization lesson. Deleting polar sectors solely because the smooth solution has zero curvature would miss an essential part of the proof.

Nor does conformality alone make a matrix model Gaussian: (4.59) permits different matter weights, and Section 4.5 gives a nontrivial conformal N=2 example. Finally, the finite-color integral and its ordered planar and strong-coupling limits are further calculations. Their asymptotic regimes are not assumptions needed for the exact localization reduction.

Exit calculations

1. Turn on the physical mass

In the local one-color example, set $\alpha=\beta=e>0$ only after forming (PC12.4). Take the stated Euclidean contour $m_P=i m_E$ and let $\mu=m_E/e$ be real. Calculate $Z_1,Z_2$. Must a nontrivial answer change a normalized Abelian Wilson loop if this factor is independent of its scalar integration variable?

Hint 1

The shifted mass is $m=e(1+i\mu)$, not $i e\mu$.

Hint 2

Keep both numerator and denominator of the normalized expectation. An overall partition-function factor can be nontrivial and still cancel.

Solution

Substitution gives

$$Z_1=-\mu^2,\qquad Z_2=\frac{\mu^4-3\mu^2}{2}. \tag{PC12.6}$$

Both vanish at zero physical mass. The coefficients need not be positive: a regulated supersymmetric fluctuation ratio is not a probability distribution over instanton numbers. For $\mu=1/2$, for example, $Z_1=-1/4$ and $Z_2=-11/32$.

The one-color formula has no Coulomb difference, so its polar factor is independent of $a$. With an insertion depending only on $a$, the same factor multiplies numerator and denominator and cancels. This matches the fact that an adjoint hypermultiplet of $U(1)$ is gauge neutral. It does not assert the same cancellation for a charged non-Abelian hypermultiplet: its weights depend on eigenvalue differences. It also does not identify the absolute partition function with its massless value. This example diagnoses precisely which statement a mass-limit calculation can establish.

2. Change the loop representation

Keep massless $SU(2)$, but use the normalized spin-one loop instead of the fundamental. For the dimensionless eigenvalue $x=\ell a_1=-\ell a_2$, the measure is proportional to $x^2e^{-A x^2}dx$, $A=16\pi^2/g^2$, and the insertion is $[1+2\cosh(4\pi x)]/3$. Evaluate the loop and check its first weak-coupling term. State what changes if the source’s unnormalized trace is retained.

Hint 1

Differentiate the ordinary Gaussian source integral twice to produce the factor $x^2$.

Hint 2

The normalized generating function is $\langle e^{w x}\rangle=(1+w^2/(2A))e^{w^2/(4A)}$. Keep the zero weight of the spin-one representation.

Solution

Writing $I(w)=\int dx\,e^{-A x^2+w x}=\sqrt{\pi/A}\,e^{w^2/(4A)}$ gives $\langle e^{w x}\rangle=I''(w)/I''(0)$, proving the stated generating function. Therefore

$$\langle W_{j=1}\rangle =\frac{1+2(1+g^2/2)e^{g^2/4}}{3} =1+\frac{g^2}{2}+O(g^4) =1+\frac\lambda4+O(\lambda^2),\qquad\lambda=2g^2. \tag{PC12.7}$$

The zero-coupling value is one. The leading term agrees with $g^2C_{j=1}/4$, since $C_{j=1}=j(j+1)=2$. An unnormalized trace multiplies the entire answer by three and tends to three at zero coupling. A missing Vandermonde would instead change the generating function itself, so it cannot be repaired by a trace normalization. This is the representation sum in Pestun (4.60), specialized to the massless adjoint theory’s trivial normal and polar factors.

After these checks, reread (1.3): identify the origin and convention of every factor, then explain exactly which two factors become one at physical zero mass. That is the useful exit from the paper into other localization problems.