2026-10-01 · chapter Yang–Mills and the mass gapPart 7 From the side of physicssupersymmetry and electric–magnetic duality

Supersymmetry and Seiberg–Witten — switching to the dual, the strongly coupled vacuum becomes a weakly coupled theory of monopoles

Pure Yang–Mills is hard because the coupling becomes strong at the low energies where the gap is decided, and perturbation theory does not work (7-01). Gauge theories with two sets of supersymmetry (N=2) are the typical neighbouring theory that can be computed into the strongly coupled side, and Seiberg and Witten (1994, SW below) proposed to fix the low-energy effective theory exactly in the case of SU(2). The abstract says 'The "strongly coupled" vacuum turns out to be a weakly coupled theory of monopoles, and with a suitable perturbation confinement is described by monopole condensation' Physics.

The three theories differ in their set of fields and in the coefficient $b$ of the one-loop running (in the convention where pure Yang–Mills has $11N/3$). One Weyl fermion in the adjoint representation (a two-component fermion; the Dirac fermion of the electron has four components) adds $-2N/3$, and one complex scalar adds $-N/3$ Known.

Example: for SU(2) these are, in order, $22/3,\ 6,\ 4$. The N=2 theory has no gap, since a massless photon remains; a gap appears only when it is broken to N=1 in §09. Section 6.6 of the problem statement also cites Seiberg–Witten as an example of a route to understanding the gap in theories with matter ('for example, see [41]') Known.

Supersymmetry is a symmetry that exchanges bosons and fermions, and N counts how many sets of its generators (supercharges) there are. A set of fields carried into one another by supersymmetry is called a multiplet. The smallest set for an N=2 gauge theory is the gauge field $A_\mu$, two Weyl fermions $\lambda,\psi$ and a complex scalar $\phi$, all in the adjoint representation (1-04). The potential of the scalar is

and it is 0 if $\phi$ and $\phi^\dagger$ commute. For SU(2) a gauge transformation brings it to $\phi=\tfrac12a\sigma^3$ ($a\in\mathbb C$), and since $a$ and $-a$ are carried into each other, the gauge-invariant coordinate is $u=\operatorname{tr}\phi^2=\tfrac12a^2$ Known. The vacua with $V=0$ form a family numbered by the complex number $u$ (a moduli space), and SW argue that this family is not lifted in the quantum theory either Physics.

For $a\neq0$ the only elements commuting with $\phi$ are the diagonal U(1), and the remaining two gauge fields (the W bosons) acquire mass by the Higgs mechanism (the mechanism by which a scalar field that takes a value gives mass to gauge fields; 3-07). The U(1) photon stays massless, so there is no gap. Example: if $\phi=\sigma^+=\begin{pmatrix}0&1\\0&0\end{pmatrix}$, then $[\phi,\phi^\dagger]=\sigma^3$ and $V=2/g^2\gt 0$; if $\phi=\tfrac12a\sigma^3$, then $V=0$.

A complex function is holomorphic if it can be differentiated as a function of a complex number. Under N=2 supersymmetry the low-energy effective theory can be written with a single holomorphic function $\mathcal F(a)$ (the prepotential) (Seiberg 1988, SW §2) Physics. The complex number combining the coupling $g$ and the θ angle,

gives the coupling of the photon and the distance (metric) on the family of vacua. Where $|a|$ is large, the theory is asymptotically free and weakly coupled (4-06). The original writes

and continues 'Higher order perturbative corrections are absent. Instantons lead to new terms.' The first term is one loop, and the $k$-th term is the contribution of $k$ instantons (3-05) Physics.

The power $4k$ comes from the $b=4$ of (1). In the convention of (3) in 7-01, with $b_0=b/16\pi^2=1/4\pi^2$,

and the exponential is the weight of a single instanton. $(\Lambda/a)^4$ appears at no order of perturbation theory (4-12) Known. Example: in the units that put the singularities at $u=\pm1$, one loop gives $\operatorname{Im}\tau\approx(2/\pi)\ln(2|a|)$ (from the numbers in §07). The coefficient $2/\pi=8\pi b_0$ agrees with the running of (5) Computed.

The metric of (3) must be positive. But the imaginary part of a holomorphic function is a harmonic function.

The value of a harmonic function at the centre of a circle equals its average over the circle, and unless it is constant it cannot have a minimum in the interior of a region (the minimum principle) Known.

If $\tau$ is a single function on the whole $a$-plane, then $\operatorname{Im}\tau$ goes to $+\infty$ as $|a|\to\infty$, so it has a minimum somewhere, contradicting the principle. The original writes 'if Im τ(a) is globally defined it cannot be positive definite as the harmonic function Im τ cannot have a minimum', and concludes that the description by $a$ is only locally correct Physics. Example: the one-loop $(2/\pi)\ln(2|a|)$ is negative for $|a|\lt1/2$, and at $|a|=1/4$ it is $-0.44$; that is, "$4\pi/g^2$ is negative" Computed. On the strongly coupled side another variable is needed.

Maxwell's equations with magnetic charges added were symmetric under exchanging electric and magnetic ((6) in 7-02). Including the θ angle, the exchange is $\tau\to-1/\tau$, and shifting θ by $2\pi$ gives $\tau\to\tau+1$; the two generate the group $SL(2,\mathbb Z)$ (integer entries, determinant 1) Known. SW set $a_D=\partial\mathcal F/\partial a$ and wrote the metric symmetrically in $a$ and $a_D$.

With $a_D$ as the coordinate the form is the same, and the coupling changes to $-1/\tau$. The original stresses that this is another description of the same theory Physics. The mass of a particle with electric charge $n_e$ and magnetic charge $n_m$ has a lower bound, and the mass $m_{\rm BPS}$ of particles equal to the bound (BPS states) is

(Witten–Olive 1978, SW §4) Physics. The W boson is $(n_m,n_e)=(0,1)$, the monopole $(1,0)$, and the dyons $(1,n)$. Example: classically $a_D=\tau_{\rm cl}a$, and the monopole is $|\tau_{\rm cl}|\approx4\pi/g^2$ times heavier than the W, about 12.6 times for $g=1$. On the strongly coupled side the ratio is reversed (Figure 2).

Differentiating the one-loop term of (4) gives $a_D\approx\frac{i}{\pi}a\big(\ln\frac{a^2}{\Lambda^2}+1\big)$. For large $u$, $a\approx\sqrt{2u}$, so when $u$ goes once round a large circle, $a\to-a$ and $\ln a^2\to\ln a^2+2\pi i$, and $a_D\to-a_D+2a$. Written as a matrix acting on the column vector $(a_D,a)^{\mathsf T}$ (the monodromy),

If every monodromy commuted with $\mathsf M_\infty$, then $a^2$ would be a global coordinate and we would be back at the contradiction of §04. SW therefore proposed that there are two singularities at finite places (choosing units, $u=\pm1$), where the mass of a monopole or a dyon becomes 0 Physics. Near $u=1$ the effect of the light monopole gives $a\approx a_0+(i/\pi)a_D\ln a_D$, and one turn gives $a\to a-2a_D$.

(The formulas of SW. In the same Figure 1, SW write the product in the order that puts the matrix of the loop traversed first on the right.) The charge $q=(n_m,n_e)$ of the particle that becomes massless satisfies $q\mathsf M=q$; for $\mathsf M_1$ it is $q=(1,0)$, that is, the monopole.

ComputedThe numbers in this figure were computed on this machine (the equation (PF) of §07 solved along the path from $P$ by the Runge–Kutta method; the matrices agree with Table 1).

In Table 1 the matrix for going round path A and then path B is written $\mathsf M_A\mathsf M_B$ (the order of analytic continuation). Seen from $P$, the large loop is "$-1$ and then $+1$", so $\mathsf M(-1)\mathsf M(1)=\mathsf M(\infty)$, and $\mathsf M(-1)=\mathsf M_\infty\mathsf M_1^{-1}=\mathsf M_\infty\mathsf M_{-1}\mathsf M_\infty^{-1}$ Computed. It is conjugate to the $\mathsf M_{-1}$ of SW, and the charge of the massless dyon changes from $(1,-1)$ to $(1,1)$. This is an instance of the original's 'complete democracy among dyons' (conjugating by $\mathsf M_\infty$ shifts the charge by 2).

SW §6 assigns to each $u$ the curve

Gluing the two sheets of $y=\pm\sqrt{\cdots}$ along two cuts joining the branch points $-1,1,u,\infty$ gives a surface with one hole (a torus). A torus is the plane divided by a lattice $\mathbb Z\omega_1+\mathbb Z\omega_2$, and with the orientation of $\tau=\omega_2/\omega_1$ chosen consistently, $\operatorname{Im}\tau>0$ Known. The branch points coincide only at $u=\pm1,\infty$, which agree with the singularities. Set the periods to be

Differentiating in $u$, the integrand becomes a constant multiple of $dx/y$, and $da/du$ and $da_D/du$ are integrals along the two loops of the torus, that is, periods of the lattice (times the same constant). $\tau$ is their ratio, so $\operatorname{Im}\tau>0$ holds automatically (the properties of the curve are Known; the correspondence with the theory is Physics). The two periods satisfy the same equation $4(u^2-1)\Pi''+\Pi=0$ (the Picard–Fuchs equation, (PF) below) (relative residual at most $10^{-15}$ on the real axis $u>1$ Computed). Figure 1 integrates this equation numerically along the path. Near $u=1$,

and the $a\to a-2a_D$ of (10) comes out. Example: the numerical integration of (12) agreed with the closed form in hypergeometric functions to within $10^{-8}$, and $a(1)=4/\pi$ Computed.

ComputedThe numbers in this figure were computed on this machine (15 points of the numerical integration of (12) embedded).

The one-loop column is $(2/\pi)\ln2a$. As $u\to1$, $\operatorname{Im}\tau$ goes to 0 and the electric coupling diverges.

Detour: at $u=3$ we get $\tau=i$

Expanding (12) at large $u$ and solving inversely for $a$ gives a series with rational coefficients $c_k$ Computed.

Matone (1995) introduced $\mathcal G=\pi i\big(\mathcal F-\tfrac12a\,a_D\big)$, and the abstract says it is modular invariant. Putting $\mathcal G=u$ into (4), the one-loop term gives $a^2/2$ and the $k$-th term gives $2\pi ik$ times the term in $a^{2-4k}$, so the coefficient of $a^{2-4k}$ in (4) reads as $c_k/(2\pi ik)$. The equation of the abstract, $(1-\mathcal G^2)\mathcal G''+\tfrac14a\,\mathcal G'^3=0$, also becomes the same equation as (PF) on setting $\mathcal G=u$ Computed.

We check this reading with $a_D$. Example: at $u=2$ the numerical integration of $a_D/i$ gives 0.473434. One loop alone gives 0.462621, up to $k=1$ 0.473083, and up to $k=3$ 0.473432; each added term shrinks the difference by more than one digit Computed. Dorey–Khoze–Mattis (1996) computed one and two instantons directly and reported 'precise agreement' Physics.

In the language of N=1, the N=2 set splits into the set $(A_\mu,\lambda)$ (the vector multiplet) and the set $\Phi$ of $\phi,\psi$. Adding a term $m\operatorname{tr}\Phi^2$ that gives mass to $\Phi$, an N=1 theory remains at low energies. The original describes this theory as 'strongly believed to generate a mass gap', and lists confinement, the breaking $\mathbb Z_4\to\mathbb Z_2$ and that there are exactly two vacua (citing Witten 1982) Physics. $\mathbb Z_4$ is the symmetry generated by the transformation that multiplies the phase of $\lambda$ by $i$, and the remaining $\mathbb Z_2$ is multiplication by $-1$.

Near $u=1$ we use the dual photon set $A_D$ and the monopole fields $M,\tilde M$ (names of fields, not masses). The holomorphic function that fixes the interactions (the superpotential) and the conditions for the vacuum are

where $U(A_D)$ is the coordinate $u$ written as a function of $A_D$ for the whole set. If $m\neq0$ and $du/da_D\neq0$, then $M,\tilde M\neq0$ is required, and $a_DM=0$ gives $a_D=0$, that is, $u=1$. From (13), $du/da_D=-2i$, so the solution with $|M|=|\tilde M|$ has $|M|^2=\sqrt2\,|m|$ (in the units of §06) Computed.

The monopoles condense. The dual photon acquires mass by the Higgs mechanism, and the original writes 'Expanding around this vacuum, it is easy to see that there is a mass gap'. What condenses is magnetic charge, so the electric field is squeezed into tubes and electric charges are confined (exactly the picture of 7-02) Physics. At $u=-1$ a dyon condenses, and there are two vacua. Example: for $m=0.01$, $|M|=0.119$ Computed. There are two scales, $\Lambda$ and $m$, so the situation differs from "a single scale" of pure Yang–Mills (7-12).

To reach pure Yang–Mills, two limits are needed: decoupling $\Phi$ with $m\to\infty$, and sending the mass of the fermion $\lambda$ to infinity as well. The weak-coupling description of §09 is for small $m$, and that the gap and confinement are not interrupted along the limits remains an expectation; there is no means to control it (7-10) Physics. The whole argument also presupposes that the N=2 theory exists. The example of SW is the typical case where, in four dimensions, confinement was derived from a duality at the level of physics (semiclassical calculations also exist, such as Polyakov's in three dimensions; 7-02), but what remained on the side of mathematics took another form.

The mathematics in Table 3 consists of theorems and definitions about integrals over the space of instantons and about solutions of equations, not claims about the spectrum (4-11). Example: the coefficients of (14), once the conventions are aligned, are numbers guaranteed by a theorem as integrals over the space of instantons. That the same numbers are "properties of the N=2 quantum field theory" cannot be said, since the construction of the theory (6-01) is missing. It is accurate to read this as an instance, at the level of physics, of the "duality transformations" of §6.6 of the problem statement (7-13, 9-02).

What this chapter can say

Sources and reproduction

Next chapter to read: 7-12 The sentence that plays the role of an equivalence principle — cutting out "a single scale" as an assumption in one sentence.

← 7-10 Back to four dimensionsContents7-12 The sentence that plays the role of an equivalence principle →