Recover a prepotential coefficient by instanton counting

The period calculation found corrections in powers of $\Lambda^4/a^4$. Can a calculation in the microscopic gauge theory recover their coefficient? We will evaluate two residues and obtain $\mathcal F_{\mathrm{inst},1}=\Lambda^4/(8\pi i a^2)$ in the period normalization already fixed there.

The other prerequisite is the general localization argument, including its contour and boundary qualifications. No three-dimensional sphere example is needed. Here the parent theory is pure four-dimensional $\mathcal N=2$ $SU(2)$ Yang–Mills: eight real supercharges, no hypermultiplets, and an Abelian rank-one low-energy theory at a generic Coulomb vacuum. The Euclidean rotational background used for the calculation preserves the chosen localization operation. It is not a claim that all eight flat-space supercharges remain manifest at nonzero deformation.

Regulate the instanton integral

An instanton is a finite-action Euclidean gauge configuration in a topological sector. Choose the anti-self-dual convention $F^+=0$ used in the primary calculation. For a Hermitian connection, ordinary fundamental trace $\operatorname{Tr}_2(t^at^b)=\delta^{ab}/2$, and the standard orientation of $\mathbb R^4$, take $k=-\int\operatorname{Tr}_2(F\wedge F)/(8\pi^2)\geq0$ in that sector. The gauge action $\int\operatorname{Tr}_2(F\wedge *F)/g^2$ is $8\pi^2k/g^2$ on such a configuration.

Its holomorphic weight is a power of an instanton parameter $q$ of mass dimension four. One can write $q=C_{\mathrm{UV}}\mu^4\exp(2\pi i\tau_{\mathrm{UV}})$, with $\tau_{\mathrm{UV}}=\theta/(2\pi)+4\pi i/g^2$ and a fixed finite normalization $C_{\mathrm{UV}}$. The curve comparison below specifies that normalization. Anti-instantons contribute to the conjugate sector; they are not another copy to add to this holomorphic prepotential.

An unregulated instanton can move arbitrarily far away. It can also shrink to zero size, where the space of configurations is singular. These are different problems. Introduce complex spacetime coordinates $z_1=x^1+ix^2$, $z_2=x^3+ix^4$ and two rotational parameters:

$$v=\epsilon_1(x^1\partial_2-x^2\partial_1) +\epsilon_2(x^3\partial_4-x^4\partial_3), \qquad [\epsilon_1]=[\epsilon_2]=1. \tag{SW16.1}$$

The Omega background combines these rotations with an R-symmetry background. In a construction with an auxiliary circle of length $\beta$, the rotations have angles $\beta\epsilon_1,\beta\epsilon_2$, accompanied by an $SU(2)_R$ rotation with eigenphases $\pm\beta(\epsilon_1+\epsilon_2)/2$. Taking the circle length to zero gives the four-dimensional construction. This explains why two independent rotations can be compatible with a selected odd symmetry; rotating spacetime alone would not establish it. See Tachikawa, (A.2.1)–(A.2.3).

In the cohomological description, which reorganizes the fields according to this symmetry, $Q^2$ is the combined rotational and gauge action. Its fixed configurations can be isolated when the parameters are generic. The rotation supplies an equivariant replacement for the divergent translation integral: a complex coordinate of weight $\epsilon_i$ contributes $1/\epsilon_i$. This is an equivariant integral, not a positive probability distribution or a literal four-volume $1/(\epsilon_1\epsilon_2)$.

A separate smooth resolution deals with zero-size instantons. We use the resolved, framed instanton construction described in Nekrasov, sections 3.1–3.4, with the geometric measure convention specified below. “Framed” means gauge transformations used in the quotient approach the identity at infinity. The asymptotic scalar parameter is held fixed; we do not integrate over Coulomb vacua as in a compact-space partition function. The source explains the resolution and the classes integrated on it. We will use a supplied one-instanton localization measure, not attempt to derive the full gauge-fixing and instanton measure here.

Evaluate the two fixed points

Let $\alpha$ denote a physical special coordinate with $\alpha\sim\sqrt u$ on the positive weak branch. The equivariant gauge eigenvalues in the convention we use are instead $b_1=b$, $b_2=-b$, where $b=-i\alpha$. This factor is explicit in Nakajima–Yoshioka’s section 2 and (2.2)–(2.3). Equivariant weights are complex parameters, and must not be silently identified with the real physical scalar on that ray. We will derive $\alpha=a/\sqrt2$ below. Set $E=\epsilon_1+\epsilon_2$.

There is no matter mass in this pure theory. Neither $E$ nor $E/2$ is an adjoint-hypermultiplet mass.

Expand the instanton partition function as

$$Z_{\mathrm{inst}}=1+qZ_1+q^2Z_2+\cdots,\qquad [Z_k]=-4k.$$

The label $k$ counts instantons, not the rank of the gauge group. We use the geometric fixed-point measure: each contribution is the inverse product of its four complex tangent weights. The two one-instanton lists are $(\epsilon_1,\epsilon_2,2b,E-2b)$ and $(\epsilon_1,\epsilon_2,-2b,E+2b)$. They are the one-box specialization of Nakajima–Yoshioka’s Theorem 3.2 and section 4.2. An equivalent residue formula, with its orientation fixed by these weights, is

$$\begin{aligned} P(\varphi)&=(\varphi-b)(\varphi+b),\\ Z_1&=-\frac{E}{\epsilon_1\epsilon_2} \oint_{\mathcal C}\frac{d\varphi}{2\pi i}\, \frac{1}{P(\varphi)P(\varphi+E)}. \end{aligned} \tag{SW16.2}$$

The variable $\varphi$ belongs to the finite-dimensional instanton quotient; it is not the spacetime scalar being integrated over all vacua. The oriented contour $\mathcal C$ surrounds $\varphi=b,-b$ counterclockwise and excludes $b-E,-b-E$. Start with distinct poles, $\epsilon_1\epsilon_2\ne0$ and $E\ne0,\pm2b$. Other nonsingular parameter values are reached by analytic continuation of the resulting rational function.

The minus sign matters. Nekrasov’s printed (3.10) uses $+E$ in this contour convention. At one instanton its coefficient is therefore $-Z_1$; translating its one-instanton generating parameter requires $q_{\mathrm{NEK}}=-q$ at fixed weights. We use the geometric Euler-class definition throughout, rather than switching signs by retaining a source’s symbol $q$ unchanged.

Why exactly two contributions? Fixed points are labeled by a Young diagram for each color. A Young diagram records a nonincreasing sequence of nonnegative row lengths; its boxes count the instanton number. At total box number one the only choices are $(\Box,\varnothing)$ and $(\varnothing,\Box)$, whose poles are $b_1$ and $b_2$. This is the $k=1$ case of the source’s classification, (3.19). It is enough here; no combinatorial formula for arbitrary diagrams is required.

For the first pole, $P'(b)=2b$ and $P(b+E)=E(2b+E)$. For the second, replace $b$ by $-b$. Thus

$$\begin{aligned} Z_1 &=-\frac{1}{\epsilon_1\epsilon_2} \left[\frac{1}{2b(2b+E)} +\frac{1}{(-2b)(-2b+E)}\right]\\ &=\frac{2}{\epsilon_1\epsilon_2(E^2-4b^2)} =\boxed{\frac{2}{\epsilon_1\epsilon_2(E^2+4\alpha^2)}}. \end{aligned} \tag{SW16.3}$$

The factor $E$ cancels the factor from the selected color in $P(b_i+E)$. The two remaining factors describe the charged directions linking the two colors; multiplying one of them by the displayed minus sign gives precisely the supplied tangent weights. Both fixed points are necessary; keeping only one gives an answer that is not invariant under the Weyl exchange $\alpha\mapsto-\alpha$.

For a worked finite-regulator example, choose a mass unit $m_0>0$ and $\alpha=m_0$, hence $b=-im_0$, $\epsilon_1=m_0/5$, $\epsilon_2=3m_0/10$. Including the overall minus sign, the two charged-direction factors are $(4-i)/(17m_0^2)$ and $(4+i)/(17m_0^2)$. Thus $Z_1=400/(51m_0^4)$ and $\epsilon_1\epsilon_2Z_1=8/(17m_0^2)$. The latter tends to $1/(2m_0^2)$ when both rotational parameters are reduced to zero. Individual fixed-point terms need not be real; the result is an equivariant sum, not a sum of probabilities.

The use of a $U(2)$ matrix construction does not change the target theory to interacting $U(2)$ Yang–Mills. It provides a convenient resolution and framing description. We impose $b_1+b_2=0$ to extract the $SU(2)$ sector. Indeed, if both eigenvalues are shifted by a common number $c$, changing integration variable $\varphi\mapsto\varphi+c$ leaves (SW16.2) unchanged. The central photon of an ordinary $U(2)$ theory is free and is not another charged instanton color to add. At this order no additional, parameter-dependent “$U(1)$ factor” is divided out. Such a division would require its own definition in a different partition function.

Extract the connected prepotential

The logarithm selects connected contributions:

$$\log Z_{\mathrm{inst}} =qZ_1+q^2\left(Z_2-\frac12Z_1^2\right)+O(q^3). \tag{SW16.4}$$

Define the connected quantity $\mathscr F_{\mathrm{inst}}=\lim_{\epsilon_1,\epsilon_2\to0} \epsilon_1\epsilon_2\log Z_{\mathrm{inst}}$, with the limit taken coefficient by coefficient in $q$, at fixed nonzero $\alpha$ and $b=-i\alpha$. The formal logarithm and this limiting prescription are defined in Nakajima–Yoshioka, section 5 opening. We keep $\mathscr F$ distinct from the physical $\mathcal F$ until translating its coordinate and its factor $2\pi i$.

At first order, no higher instanton sector enters:

$$\mathscr F_{\mathrm{inst}} =\frac{q}{2\alpha^2} +O\!\left(\frac{q^2}{\alpha^6}\right). \tag{SW16.5}$$

Its mass dimension is two. For $\epsilon_1=t$, $\epsilon_2=ct$ with fixed nonzero $c$, the regulated coefficient is

$$\epsilon_1\epsilon_2 Z_1 =\frac{1}{2\alpha^2} \left[1-\frac{(1+c)^2t^2}{4\alpha^2} +O\!\left(\frac{t^4}{\alpha^4}\right)\right]. \tag{SW16.6}$$

This converges uniformly on parameter sets bounded away from $\alpha=0$ with bounded $c$. The special path $\epsilon_2=-\epsilon_1$ gives the limiting coefficient exactly after evaluating (SW16.3). Setting $E=0$ in the unevaluated contour instead would merge selected and excluded poles and mistakenly turn a finite limit into zero.

The regulator poles $E=\pm2b=\mp2i\alpha$ signal a failure of the isolated-weight formula there. They are not the two physical strong-coupling singularities of the undeformed theory; for real nonzero $\alpha$ and real $E$ they are not encountered at all. Likewise, this weak-coupling expansion must not be extrapolated to $\alpha=0$. Its physical small parameter is $|q/\alpha^4|$.

Translate the curve and the electric coordinate

We must match scales before comparing a number. In the instanton-source convention, the pure rank-one spectral curve and its differential are

$$w+\frac q w=X^2-u,\qquad \lambda_{\mathrm{I}}=X\,\frac{dw}{w},\qquad \alpha=\frac{1}{2\pi i}\oint_A\lambda_{\mathrm{I}}. \tag{SW16.7}$$

Setting $w=\sqrt q\,z$ gives Tachikawa’s (A.1.1)–(A.1.2). The geometric counting convention identifies $q=\Lambda_{\mathrm I}^4$; see Nakajima–Yoshioka, section 4.4. Its differential is $-\lambda_{\mathrm I}/(2\pi)$, so its electric period is $b=-i\alpha$, exactly the continuation used above. The polynomial is $X^2-u$; the same source calls its constant coefficient $u_2=-u$. The symbol $u$ here remains the invariant of the preceding lessons. The electric contour has positive $\alpha\sim\sqrt u$ on the real weak branch.

The branch points in the $X$ plane obey $X^2=u\pm2\sqrt q$. Two collide at $u=\pm2\sqrt q$. Matching these degeneration values to the already fixed $u=\pm\Lambda^2$ gives $q=\Lambda^4/4$, with the positive scale branch chosen on the real weak ray. This fixes a finite instanton-scale convention, including its phase; it is not permission to adjust the scale separately for each coefficient.

The coordinate factor can also be checked without fitting the coefficient. On $z=e^{i\vartheta}$, for real $u>2\sqrt q>0$,

$$\alpha(u)=\frac{1}{2\pi}\int_0^{2\pi} \sqrt{u+2\sqrt q\cos\vartheta}\,d\vartheta =\frac{a(u)}{\sqrt2}. \tag{SW16.8}$$

The last equality follows by shifting the angle by $\pi$ and comparing with the electric integral in the period lesson. It is an equality of special coordinates on this branch, continued consistently elsewhere. It is not the incorrect quantum identity $\alpha^2=u$.

For the instanton part, the complete translation we use is therefore

Quantity in the instanton formula	Quantity in the fixed period convention
$\alpha$	$a/\sqrt2$
Equivariant eigenvalue $b$	$-ia/\sqrt2$
$q=\Lambda_{\mathrm I}^4$	$\Lambda^4/4$
$\mathscr F_{\mathrm{inst}}$	$2\pi i\,\mathcal F_{\mathrm{inst}}$ after the coordinate substitution
$\partial_\alpha$	$\sqrt2\,\partial_a$

The two curve presentations also have different integral period lattices. Write the instanton curve as $Y^2=(X^2-u)^2-4q$. When $4q=\Lambda^4$, the map $x=u-X^2$, $y=iXY$ sends it to the site’s cubic $y^2=(x^2-\Lambda^4)(x-u)$. Generically two source points map to one target point. On the chosen weak branch, the electric cycle maps to the site’s electric cycle, while the magnetic cycle maps to twice the site’s magnetic cycle. The differential pulls back to $\sqrt2\lambda_{\mathrm I}/(2\pi i)$. Consequently, with $\alpha_D=(2\pi i)^{-1}\oint_{B_{\mathrm I}}\lambda_{\mathrm I}$,

$$a=\sqrt2\alpha,\qquad \alpha_D=\sqrt2a_D,\qquad \tau_{\mathrm I}=\frac{d\alpha_D}{d\alpha}=2\tau.$$

This is why electric rescaling alone is not enough to identify the coupling. The prepotential factor follows from the same source’s (2.4): its period matrix is $-(2\pi i)^{-1}\partial_b^2\mathscr F$. With $b=-i\alpha$, this becomes $(2\pi i)^{-1}\partial_\alpha^2\mathscr F=\tau_{\mathrm I}$. Thus the physical prepotential in the $\alpha$ coordinate is $\mathscr F/(2\pi i)$. The cycle factors above agree with $\mathcal F_\alpha(\alpha)=\mathcal F_{\mathrm{site}}(\sqrt2\alpha)$, so substituting $\alpha=a/\sqrt2$ gives the site’s normalization.

These statements concern the instanton part. The perturbative prepotential also has a finite quadratic term fixed by the earlier period branch and scale. It must be matched separately, rather than silently imported from a source that suppresses such a term. Nor is this electric-unit translation an integral change of the site’s charge basis.

Using (SW16.5),

$$\begin{aligned} \mathcal F_{\mathrm{inst},1}(a) &=\frac{1}{2\pi i}\, \frac{\Lambda^4/4}{2(a/\sqrt2)^2} =\boxed{\frac{\Lambda^4}{8\pi i a^2}},\\ a_{D,\mathrm{inst},1} &=\frac{i\Lambda^4}{4\pi a^3},\qquad \tau_{\mathrm{inst},1} =-\frac{3i\Lambda^4}{4\pi a^4}. \end{aligned} \tag{SW16.9}$$

The derivative factors matter: differentiating with respect to $\alpha$ and calling the result $a_D$ would give the wrong normalization.

There is a useful source-reading caution. With the same real symbol $a$ interpreted literally, Tachikawa v3 (A.2.22) has the opposite zero-rotation sign from the positive coefficient in the following (A.2.23). Keeping the geometric equivariant coordinate $b$ distinct from the physical $\alpha$, as above, removes this ambiguity. The resulting physical coefficient agrees with the curve value $d_1=1/2$ in (A.1.15). A measure, an instanton parameter and a prepotential convention must be translated together.

Recover the same coefficient from the periods

Let $L=\log(8u/\Lambda^2)$. The previously derived weak-branch expansions are $a=\sqrt{2u}[1-\Lambda^4/(16u^2)+\cdots]$ and $a_D=i\sqrt{2u}[L-2-\Lambda^4(L-1)/(16u^2)+\cdots]/\pi$. First express both in the same independent coordinate:

$$\begin{aligned} u&=\frac{a^2}{2}+\frac{\Lambda^4}{4a^2}+O(\Lambda^8/a^6),\\ \sqrt{2u}&=a\left[1+\frac{\Lambda^4}{4a^4} +O(\Lambda^8/a^8)\right],\\ L&=\log\frac{4a^2}{\Lambda^2} +\frac{\Lambda^4}{2a^4}+O(\Lambda^8/a^8). \end{aligned} \tag{SW16.10}$$

Substituting into $a_D$ cancels the logarithm multiplying $\Lambda^4/a^4$. The remaining coefficient is $1/4$:

$$a_D(a)=\frac{ia}{\pi} \left[\log\frac{4a^2}{\Lambda^2}-2\right] +\frac{i\Lambda^4}{4\pi a^3} +O\!\left(\frac{\Lambda^8}{a^7}\log\frac a\Lambda\right). \tag{SW16.11}$$

Integrating the correction with respect to $a$ gives exactly (SW16.9). The leading term integrates to $\mathcal F_{\mathrm{pert}}=ia^2[\log(4a^2/\Lambda^2)-3]/(2\pi)$, up to an irrelevant additive constant. Thus the comparison retains the finite scale convention instead of comparing only powers of $a$.

A practical numerical diagnostic is $C(a)=4\pi a^3[a_D(a)-a_{D,\mathrm{pert}}(a)]/(i\Lambda^4)$. Evaluate $a,a_D$ by the independent contour integrals, then increase $u/\Lambda^2$ on the real weak branch. The prediction is $C\to1$; subtracting the perturbative term needs higher arithmetic precision as the correction becomes small. One can also integrate (SW16.2) around small circles and compare each result with its residue. These checks test different operations: the microscopic contour selection and the normalization of the macroscopic periods.

For example, contour quadrature gives $C=1.0018454982$, $1.0004586599$, $1.0001144968$ and $1.0000286137$ at $u/\Lambda^2=8,16,32,64$. These are finite-distance checks of the approach to one, not four estimates of an exact equality at finite $a$.

This is a first-coefficient test of the two-derivative effective theory in a specified scheme. It does not derive every instanton sector, the full nonperturbative identification of partition functions, or the stable BPS spectrum.

Exercises

1. Compare two regulator limits

Take $\alpha\ne0$. Evaluate $\epsilon_1\epsilon_2Z_1$ along $\epsilon_2=-\epsilon_1$, and compare with the limit $\epsilon_2\to0$ at fixed $\epsilon_1=\hbar\ne0$. What extra limit is required in the second case? Locate the latter expression’s poles for complex $\hbar$ and explain why they are not singularities of the undeformed Coulomb branch.

Hint 1

Use the evaluated rational function, not an unevaluated contour with colliding poles.

Hint 2

The second prescription retains a rotational scale. Compare it with the assumptions used to identify a flat-space prepotential.

Solution

For $\epsilon_2=-\epsilon_1$, $E=0$ and $\epsilon_1\epsilon_2Z_1=1/(2\alpha^2)$ exactly, by analytic continuation from generic parameters. The original contour poles merge when $E=0$, so substituting before evaluating is invalid.

At fixed $\hbar$,

$$\lim_{\epsilon_2\to0}\epsilon_1\epsilon_2Z_1 =\frac{2}{4\alpha^2+\hbar^2} =\frac{1}{2\alpha^2} \left[1-\frac{\hbar^2}{4\alpha^2}+\cdots\right].$$

Only after $\hbar\to0$ at fixed nonzero $\alpha$ does this become the flat coefficient. A one-parameter limit retains deformation-dependent information. Its poles at $\hbar=\pm2i\alpha$ are vanishing equivariant weights; they depend on an external regulator and are distinct from $u=\pm\Lambda^2$. Both procedures give an object of mass dimension minus two after removing the translation factor.

2. Diagnose the contour and the gauge center

Replace $\mathcal C$ by a counterclockwise contour enclosing all four poles of (SW16.2). Determine the result and explain why it is not an alternative computation of the same instanton integral. Then replace the equivariant eigenvalues by $c+b,c-b$ and show explicitly that the correct answer is independent of $c$.

Hint 1

Inspect the integrand at infinity. The residue there is minus the coefficient of $1/\varphi$.

Hint 2

For the second part, shift the integration variable and the contour together. Keep the distinction between a quotient coordinate and a spacetime vacuum parameter.

Solution

The rational integrand falls as $\varphi^{-4}$, so it has no residue at infinity. The sum of all four finite residues is zero. Equivalently, the excluded poles contribute the negative of the selected sum:

$$-\frac{E}{\epsilon_1\epsilon_2} \sum_{\varphi=b-E,-b-E} \operatorname{Res}\frac{1}{P(\varphi)P(\varphi+E)} =-Z_1.$$

The original contour encodes the resolved quotient prescription. Changing which poles it encloses changes that prescription; Cauchy’s theorem does not allow crossing poles without adding their residues.

With $P_c(\varphi)=(\varphi-c)^2-b^2$, put $\psi=\varphi-c$. Then $P_c(\varphi)=P(\psi)$ and $P_c(\varphi+E)=P(\psi+E)$; the selected poles become $\psi=\pm b$. Hence $Z_1$ is unchanged. Imposing $c=0$ selects the traceless boundary data. This demonstrates center independence of the specified coefficient, not a theorem that every $U(2)$ partition function equals an $SU(2)$ one.

3. Why the logarithm must be expanded first

Suppose one retains only $1+qZ_1$ and sends both $\epsilon_i$ to zero at a fixed positive $q$, on a real path where $qZ_1>0$. Show that this gives the wrong answer. More generally, along a fixed regulator path write $Z_1=z/\delta+z_0+O(\delta)$ and $Z_2=A/\delta^2+B/\delta+O(1)$, where $\delta=\epsilon_1\epsilon_2$ and $z,z_0,A,B$ are the path’s expansion coefficients, independent of $\delta$. What relation is needed for a finite connected two-instanton contribution, and what is that contribution? What would multiplying $Z_{\mathrm{inst}}$ by $\exp[q\,h(\alpha)/\delta]$ do?

Hint 1

Compare $\delta\log(1+qz/\delta)$ with the coefficient of $q$ in that same expression.

Hint 2

Use (SW16.4). A normalization depending on the Coulomb parameter can affect derivatives of the prepotential.

Solution

For $\delta\to0^+$, $\delta\log(1+qz/\delta)\to0$ if $qz>0$. But its coefficient of $q$ is $z$, which is $1/(2\alpha^2)$ in our problem. The truncation is not uniform: $qZ_1$ diverges at fixed $q$ as the regulator is removed. A calculation of one formal coefficient requires expanding in $q$ first. It cannot replace the full partition function at fixed $q$.

At the next order,

$$\delta\left(Z_2-\frac12Z_1^2\right) =\frac{A-z^2/2}{\delta}+B-zz_0+O(\delta).$$

A finite limit requires $A=z^2/2$. This is the subtraction of the disconnected pair; the connected coefficient is then $B-zz_0$. Even a finite correction to $Z_1$ can contribute to that subtraction. For $\epsilon_1=t$, $\epsilon_2=ct$ our formula gives $z=1/(2\alpha^2)$ and $z_0=-(1+c)^2/(8c\alpha^4)$. The full connected limit is independent of the path after the appropriate terms in $Z_2$ are included. Knowing only $Z_1$ does not determine it.

Finally, the proposed multiplier adds $q\,h(\alpha)$ to $\mathscr F_{\mathrm{inst}}$. If $h$ depends on $\alpha$, it generally changes the dual period and the effective coupling after the coordinate translation. It is not an innocuous partition-function normalization. An $\alpha$-independent addition would not change those derivatives, but still belongs to the specified generating functional. This is why no unmotivated center factor was removed above.

With the first coefficient under control, the geometric fixed-point formula poses a concrete next calculation: enumerate the five two-box color assignments, form the connected combination, and test the next period coefficient using the same scale and equivariant coordinate. The new task is the next instanton sector, not a new choice of normalization.

Continue to instantons, quantum periods, and differential equations to retain one rotational deformation, match the first coefficient to a specified spectral problem, and resolve its first resonance.