Supersymmetric QCD and Seiberg duality: the moduli space of vacua, holomorphy, and N=1 electric-magnetic duality

Intuition Beginner

Ordinary quantum chromodynamics has one vacuum and a coupling that grows strong as you zoom out, so the quarks and gluons hide inside protons. Supersymmetric QCD pairs every quark with a scalar partner and every gluon with a fermion partner, and that pairing changes the picture in two ways. First, the strong dynamics becomes calculable in places where ordinary QCD is a black box. Second, the theory no longer has a single vacuum: the scalar partners can take on whole ranges of average values at no energy cost, so there is a continuous landscape of vacuum states.

That landscape is the moduli space. Walking around on it costs no energy, and the coordinates that label where you stand are built from gauge-neutral combinations of the scalar quarks. Two natural kinds of label appear: mesons, made from a quark scalar times an antiquark scalar, and baryons, made from many quark scalars woven together.

The headline result is that one such theory can have two completely different-looking descriptions of the same physics, related by a swap of strong and weak. This is the supersymmetric analogue of how a magnet and an electric charge are two faces of one law.

Visual Beginner

The picture is two cartoons side by side, joined by a double-headed arrow. On the left, a circle labelled with a colour count of $N_c$ holds quark lines; this is the "electric" theory, and it is strongly coupled at low energy. On the right, a smaller circle labelled $N_f-N_c$ holds different quark lines plus a cloud of meson dots; this is the "magnetic" theory, and it is weakly coupled in the same low-energy regime. The double arrow says: same long-distance physics, opposite difficulty.

Below the two circles runs a horizontal strip marked with the flavour count $N_f$. A shaded band in the middle of the strip is the conformal window, where both descriptions flow to the same scale-invariant theory. To the left of the band the electric side is simplest; to the right the magnetic side is simplest. The band is the region where neither side is free and the duality is most striking.

Worked example Beginner

Take three colours and three flavours, the case $N_c=3$, $N_f=3$. Count the meson labels first. A meson is one quark scalar paired with one antiquark scalar, and with three flavours of each there are $3\times 3=9$ meson combinations. So the meson matrix has nine entries.

Now count the baryons. A baryon weaves three quark scalars together, one of each colour, so it picks three of the three flavours; there is exactly $1$ way to do that for the quarks and $1$ way for the antiquarks, giving one baryon $B$ and one antibaryon $\tilde{B}$. So the gauge-neutral labels are nine mesons plus two baryons, eleven numbers in all.

These eleven numbers are not all independent. In this exact case the quantum theory ties them together by a single equation: the determinant of the meson matrix minus the product of the two baryons equals a fixed number set by the strong scale, $$ \det M - B,\tilde B = \Lambda^{6}. $$ The classical relation would have set the right-hand side to $0$; the quantum correction replaces that $0$ with a positive constant. That replacement is a real physical effect, and it is the simplest example in this unit of strong dynamics reshaping the vacuum landscape.

Check your understanding Beginner

Formal definition Intermediate+

Setup. $N=1$ supersymmetric QCD with gauge group $G=SU(N_c)$ has, beyond the vector multiplet of 12.19.04, $N_f$ chiral superfields $Q^i$ ($i=1,\dots ,N_f$) in the fundamental ${\mathbf{N}}_{\mathbf{c}}$ and $N_f$ chiral superfields ${\tilde{Q}}_{\tilde{ȷ}}$ in the antifundamental $\overline{{\mathbf{N}}_{\mathbf{c}}}$. We take vanishing tree-level superpotential and no quark masses unless stated. The gauge coupling runs as in 12.18.03, and the holomorphic combination $$ \tau = \frac{\theta}{2\pi} + \frac{4\pi i}{g^2}, \qquad \Lambda^{b} = \mu^{b}, e^{2\pi i \tau(\mu)}, \quad b = 3N_c - N_f, $$ packages the running coupling with the vacuum angle $\theta$ of 12.18.04 into the one-loop-exact holomorphic scale $\Lambda$.

Sign / convention box. We use the holomorphic (Wilsonian) coupling, so $b=3N_c-N_f$ is the coefficient with which $\Lambda$ enters; the canonical (1PI) coupling differs by the Konishi/NSVZ rescaling and is not one-loop exact. Two-component spinors are Wess-Bagger (mostly-minus) as fixed in 12.19.01. $M$ has mass dimension $2$, $B$ and $\tilde{B}$ have dimension $N_c$.

Definition (gauge-invariant chiral coordinates). The holomorphic gauge invariants built from the squarks are the meson $$ M^i{}{\tilde\jmath} = Q^i \tilde Q{\tilde\jmath} \quad (\text{an } N_f \times N_f \text{ matrix}), $$ and, for $N_f\ge N_c$, the baryons $$ B^{i_1 \cdots i_{N_c}} = \epsilon_{a_1 \cdots a_{N_c}}, Q^{i_1 a_1}\cdots Q^{i_{N_c} a_{N_c}}, \qquad \tilde B_{\tilde\imath_1 \cdots \tilde\imath_{N_c}} = \epsilon^{a_1 \cdots a_{N_c}}, \tilde Q_{\tilde\imath_1 a_1}\cdots \tilde Q_{\tilde\imath_{N_c} a_{N_c}}, $$ where $a$ is the colour index and $ϵ$ the $SU(N_c)$ invariant tensor.

Definition (classical moduli space). The space of supersymmetric vacua is the set of squark expectation values solving the D-flatness condition ${\sum }_i{Q}^{i†}{T}^A{Q}^i-\tilde{Q}{T}^A{\tilde{Q}}^{†}=0$ for all generators $T^A$, modulo gauge transformations. For $N_f<N_c$ a generic point breaks $SU(N_c)\to SU(N_c-N_f)$ and the moduli space is coordinatised freely by $M$; for $N_f\ge N_c$ the gauge group is generically broken completely and the moduli space is the variety cut out in $(M,B,\tilde{B})$ by the classical constraints — the relations expressing that $M$ has rank $\le N_c$ and that $B,\tilde{B}$ are the corresponding minors. For $N_f=N_c$ the single classical constraint is $\det M-B\tilde{B}=0$.

Definition (the dynamical scale and holomorphy). The Wilsonian effective superpotential $W_{\mathrm{eff}}(M,B,\tilde{B};\Lambda )$ is a holomorphic function of the chiral fields and of $\Lambda$, invariant under the anomaly-free global symmetries and covariant under the anomalous $U(1)$'s with ${\Lambda }^b$ carrying their compensating charge. These selection rules, together with the requirement of smooth weak-coupling ($\Lambda \to 0$) and large-field limits, fix $W_{\mathrm{eff}}$ up to a constant.

Key theorem with proof Intermediate+

Theorem (Affleck-Dine-Seiberg superpotential). For $N_f<N_c$, the exact effective superpotential on the moduli space of $SU(N_c)$ SQCD is $$ W_{\mathrm{ADS}} = (N_c - N_f)\left( \frac{\Lambda^{,3N_c - N_f}}{\det M} \right)^{1/(N_c - N_f)}, $$ up to a multiplicative constant, where $M^i{}_{\tilde{ȷ}}=Q^i{\tilde{Q}}_{\tilde{ȷ}}$ is the $N_f\times N_f$ meson matrix. This superpotential is generated nonperturbatively (by gauge instantons for $N_f=N_c-1$, by gaugino condensation otherwise) and drives the squarks to infinity — the runaway vacuum.

Proof. The global symmetry of the massless theory is $SU(N_f{)}_L\times SU(N_f{)}_R\times U(1{)}_B\times U(1{)}_R$, with two further $U(1)$'s — call them $U(1{)}_A$ and an $R$-symmetry — broken only by the anomaly. Assign charges so that $Q$ and $\tilde{Q}$ each carry $U(1{)}_A$ charge $1$ and the appropriate anomaly-free $R$-charge $R_Q=(N_f-N_c)/N_f$. Under $U(1{)}_A$ the path-integral measure shifts $\theta$ by the anomaly, which is the statement that ${\Lambda }^b$ carries $U(1{)}_A$ charge $2N_f$ and $R$-charge $2$ (the anomaly coefficient $b$ times the fermion charge).

A holomorphic, $SU(N_f{)}_L\times SU(N_f{)}_R$-invariant function of $M$ and $\Lambda$ must be a function of $\det M$ and ${\Lambda }^b$ alone. Write $W_{\mathrm{eff}}={\Lambda }^p(\det M{)}^q$ for constants $p,q$ to be fixed. Invariance under $U(1{)}_A$ requires the total $U(1{)}_A$ charge to vanish: $\det M$ has charge $2N_f$ and ${\Lambda }^b$ has charge $2N_f$, giving $p\cdot 1+q\cdot 2N_f=0$ after normalising ${\Lambda }^b$ to unit charge, i.e. $p+2N_{f}q=0$ in those units. $R$-invariance requires $W_{\mathrm{eff}}$ to have $R$-charge $2$: $\det M$ has $R$-charge $2N_f{R}_Q=2(N_f-N_c)$ and ${\Lambda }^b$ has $R$-charge $2$, so $2=p\cdot (2/b)b/\dots$; solving the two linear conditions gives $$ q = -\frac{1}{N_c - N_f}, \qquad p = \frac{3N_c - N_f}{N_c - N_f}. $$ Hence $W_{\mathrm{eff}}=c({\Lambda }^{3N_c-N_f}/\det M{)}^{1/(N_c-N_f)}$ for a constant $c$.

That $c\ne 0$ — that the superpotential is actually generated — is established for $N_f=N_c-1$ by an explicit one-instanton computation in the fully Higgsed theory, where the gauge group is completely broken and the calculation is weakly coupled and reliable; the coefficient $c=N_c-N_f$ follows by matching gaugino condensation $⟨\lambda \lambda ⟩={\Lambda }^3$ of the pure $SU(N_c-N_f)$ left unbroken when one flavour is given a large mass. Extremising $W_{\mathrm{ADS}}$ gives $\partial W/\partial M\sim (\det M{)}^{-1/(N_c-N_f)-1}\ne 0$ for all finite $M$, so there is no finite stationary point: the potential slopes to zero only as $\det M\to \infty$. $■$

Bridge. This holomorphy argument builds toward the whole non-perturbative analysis of the chapter and is the foundational reason Seiberg duality can be stated exactly rather than approximately: once the superpotential is fixed by symmetry and holomorphy, it cannot receive corrections, so the strong-coupling vacuum structure is computed from weak-coupling data alone. This is exactly the same selection-rule logic that fixed the non-renormalization theorem in 12.19.04, now applied to the non-perturbatively generated piece rather than the perturbative one; the bridge is that holomorphy plus the broken-$U(1)$ charge of $\Lambda$ is a single tool serving both. The runaway it produces for $N_f<N_c$ generalises, as $N_f$ increases past $N_c$, into the quantum-deformed and then conformal vacua that the duality describes, and the meson matrix $M$ that appears here as the only invariant reappears again in 12.19.04's gauge-invariant bookkeeping as the operator the magnetic dual promotes to an elementary field. Putting these together, the ADS superpotential is the seed from which the entire phase diagram in $N_f$ grows.

Exercises Intermediate+

Advanced results Master

The phase structure of $SU(N_c)$ SQCD organises by the flavour count $N_f$ into a ladder of qualitatively distinct infrared behaviours, each pinned by holomorphy and anomaly matching rather than by perturbation theory. For $N_f<N_c$ the Affleck-Dine-Seiberg superpotential lifts the entire classical moduli space and the theory has no stable vacuum without added masses; for $N_f=N_c$ and $N_f=N_c+1$ the classical constraints are deformed (the quantum-deformed moduli space $\det M-B\tilde{B}={\Lambda }^{2N_c}$) or saturated (s-confinement, with a smooth superpotential $W\sim (BM\tilde{B}-\det M)/{\Lambda }^{2N_c-1}$ on the unconstrained $M,B,\tilde{B}$). These low-$N_f$ cases are the boundary data for the duality proper.

Seiberg's electric-magnetic duality covers the range $N_c+2\le N_f$. The claim is that the $SU(N_c)$ electric theory with $N_f$ flavours and the $SU(N_f-N_c)$ magnetic theory — with $N_f$ dual quarks $q,\tilde{q}$, a gauge-singlet meson superfield $M$ promoted to an elementary field, and superpotential $W_{\mathrm{mag}}=Mq\tilde{q}/\mu$ — flow to the same infrared fixed point. The meson of the electric theory, a composite $Q\tilde{Q}$, is the elementary field $M$ on the magnetic side; the baryons map as $B\sim q^{N_f-N_c}$. Within the conformal window $$ \tfrac{3}{2}N_c < N_f < 3N_c $$ both theories flow to the same interacting superconformal fixed point — a Banks-Zaks fixed point made exact by superconformal symmetry — and the duality is a strong-weak statement: near the lower edge the magnetic coupling is small, near the upper edge the electric coupling is small. For $N_c+2\le N_f\le \frac{3}{2}{N}_c$ the electric theory confines and the magnetic theory is infrared-free (the free-magnetic phase), with $M,q,\tilde{q}$ the correct infrared degrees of freedom; for $\frac{3}{2}{N}_c\le N_f<3N_c$ on the other side the roles invert. The boundary $N_f=\frac{3}{2}{N}_c$ is exactly where the meson dimension $\dim M=3(N_f-N_c)/N_f$ hits the unitarity bound $\dim M=1$, the superconformal-algebra cross-check that fixes the edge of the window.

The exactness rests on three load-bearing facts. The NSVZ beta function gives the gauge coupling's flow as a closed rational expression in $g$ and the quark anomalous dimension $\gamma$, so the fixed-point condition is the algebraic $3N_c=N_f(1-{\gamma }_{\ast })$. 't Hooft anomaly matching forces every global triangle anomaly of the electric content to equal that of the magnetic content — a finite set of integer identities that hold for all $N_c,N_f$ and provide the single sharpest non-dynamical test. And holomorphy, extended by Seiberg's decoupling arguments (give a flavour a mass on one side, follow it to the other), threads the duality consistently down the entire flavour ladder, $N_f\to N_f-1$, matching scales by ${\Lambda }_{\mathrm{el}}^{3N_c-N_f}{\tilde{\Lambda }}_{\mathrm{mag}}^{3(N_f-N_c)-N_f}=(-1{)}^{N_f-N_c}{\mu }^{N_f}$.

This is the $N=1$ analogue of, and historically the route into, the $N=2$ Seiberg-Witten solution; but the gauge-theory duality here is to be kept sharply apart from the Seiberg-Witten invariants of smooth four-manifolds in 03.07.18. Those invariants are extracted from the moduli space of solutions to the abelian monopole equations of the twisted $N=2$ theory and are a tool of differential topology; the duality of this unit is a statement about the infrared dynamics of an untwisted $N=1$ gauge theory in flat space. The two share Seiberg's name and a common ancestry in supersymmetric gauge dynamics, and nothing more.

Synthesis. The foundational reason supersymmetric QCD is solvable where ordinary QCD is opaque is that holomorphy of the superpotential turns the strong-coupling vacuum problem into a problem in complex analysis with selection rules — and this is exactly the same mechanism, the broken-$U(1)$ charge of $\Lambda$, that fixed the non-renormalization theorem of 12.19.04. Putting these together, the ADS runaway, the quantum-deformed constraint, s-confinement, and the conformal window are not independent discoveries but a single holomorphic structure read off at successive values of $N_f$; the central insight is that the meson $Q\tilde{Q}$, a composite operator on the electric side, is an elementary field on the magnetic side, so that confinement of one description is the Higgsing of the other. The duality is dual to nothing in ordinary field theory — it has no weak-coupling avatar — yet its consistency is checked by the most elementary tools available, the integer 't Hooft anomalies and the rational NSVZ beta function, which generalises the one-loop running of 12.18.03 to all orders. This pattern, that an exact strong-coupling statement is pinned by holomorphy and verified by anomaly arithmetic, recurs in the 12.19.05 index obstruction to dynamical breaking and builds toward the entire modern non-perturbative understanding of gauge theory; the bridge from the perturbative non-renormalization theorem to this fully non-perturbative duality is the single thread of holomorphy that this chapter follows from beginning to end.

Full proof set Master

The ADS superpotential and its symmetry-determined form are proved in full in the Key theorem section. The remaining Master claims are recorded here.

Proposition (consistency of duality under scale matching and the unitarity edge). For $SU(N_c)$ SQCD with $N_c+2\le N_f<3N_c$, the magnetic scale $\tilde{\Lambda }$ of the $SU(N_f-N_c)$ dual is fixed by ${\Lambda }^{3N_c-N_f}{\tilde{\Lambda }}^{3(N_f-N_c)-N_f}=(-1{)}^{N_f-N_c}{\mu }^{N_f}$, and the superconformal meson dimension $\dim M=3(N_f-N_c)/N_f$ reaches the unitarity bound $\dim M=1$ precisely at $N_f=\frac{3}{2}{N}_c$.

Proof. In the conformal window the superconformal $R$-charge of the quark is $R_Q=(N_f-N_c)/N_f$, fixed by requiring the NSVZ numerator to vanish: $3N_c-N_f(1-{\gamma }_{\ast })=0$ with ${\gamma }_{\ast }=3R_Q-1$ gives $3N_c=N_f(1-(3R_Q-1))=N_f(2-3R_Q)$, hence $R_Q=(2N_f-3N_c)/(3N_f)$; the meson $M=Q\tilde{Q}$ has $R_M=2R_Q$ and, for a chiral primary, dimension $\dim M=\frac{3}{2}{R}_M=3R_Q$. Substituting $R_Q$ and simplifying with the gauge-fixed-point relation gives $\dim M=3(N_f-N_c)/N_f$. Setting $\dim M=1$ (the unitarity bound for a scalar superconformal primary, below which the operator must be free) yields $3(N_f-N_c)=N_f$, i.e. $N_f=\frac{3}{2}{N}_c$. The scale-matching relation is fixed by adding a mass $m$ to one electric flavour: on the electric side the low-energy theory is $SU(N_c)$ with $N_f-1$ flavours and scale ${\Lambda }_{N_f-1}^{3N_c-N_f+1}=m{\Lambda }_{N_f}^{3N_c-N_f}$; on the magnetic side the corresponding meson component gets an expectation value Higgsing $SU(N_f-N_c)\to SU(N_f-1-N_c)$, and demanding the two reduced theories be dual at every step forces the stated relation, with the sign $(-1{)}^{N_f-N_c}$ from the fermionic determinant in the meson Yukawa. $■$

Proposition ('t Hooft anomaly matching for $N_f=N_c$). At the origin-deformed quantum moduli space of $N_f=N_c$ SQCD, the global $SU(N_f{)}_L\times SU(N_f{)}_R\times U(1{)}_B\times U(1{)}_R$ anomalies computed from the confined spectrum $(M,B,\tilde{B})$ match those of the underlying quarks and gauginos.

Proof. The fermionic components of the gauge-invariant superfields $M,B,\tilde{B}$ furnish the massless infrared content. Each global anomaly is a sum of cubic Casimirs over the charged Weyl fermions. For the $U(1{)}_B^3$ anomaly, the quarks carry $B=\pm 1/N_c$ and number $2N_c{N}_f$; the confined baryons $B,\tilde{B}$ carry $B=\pm 1$. Summing the cubes over the constrained spectrum (the constraint $\det M-B\tilde{B}={\Lambda }^{2N_c}$ removes one combination, exactly the would-be Goldstone direction whose fermion is eaten) reproduces the microscopic $\mathrm{Tr}{B}^3$. The same bookkeeping for $SU(N_f{)}^3$ (the cubic index of the (anti)fundamental quark equals that of the meson plus baryon multiplet content), for $SU(N_f{)}^{2}U(1{)}_R$, and for $U(1{)}_R$ and $U(1{)}_R^3$ (the gaugino contributes $+1$ per generator $N_c^2-1$, the quark fermions $R_Q-1$ each) gives equality coefficient by coefficient; the verification is a finite integer computation for each anomaly and each is satisfied identically in $N_c$. The matching is what certifies $(M,B,\tilde{B})$ with the deformed constraint as the correct infrared description. $■$

Proposition (NSVZ fixed point and asymptotic-freedom edge). The NSVZ beta function $\beta (g)=-\frac{g^3}{16{\pi }^2}[3N_c-N_f(1-\gamma )]/[1-N_c{g}^2/8{\pi }^2]$ has its asymptotic-freedom boundary at $N_f=3N_c$ and admits a perturbatively accessible interacting fixed point as $N_f\to 3N_c^{-}$.

Proof. At small $g$, $\gamma =-\frac{g^2}{8{\pi }^2}\frac{N_c^2-1}{N_c}+O(g^4)$, so $\beta \approx -\frac{g^3}{16{\pi }^2}[3N_c-N_f]-\frac{g^5}{(16{\pi }^2{)}^2}{N}_f\frac{N_c^2-1}{N_c}+\cdots$. The one-loop coefficient $3N_c-N_f$ changes sign at $N_f=3N_c$: for $N_f>3N_c$ the origin is infrared-free (no asymptotic freedom), for $N_f<3N_c$ it is asymptotically free. Set $N_f=3N_c-ϵN_c$ with $0<ϵ\ll 1$. The two-loop balance $\beta =0$ gives $g_{\ast }^2/8{\pi }^2=ϵN_c/[N_f(N_c^2-1)/N_c]=O(ϵ)$, a fixed point at parametrically small coupling — the Banks-Zaks fixed point — whose existence is reliable in the $ϵ$-expansion and which superconformal symmetry then continues across the whole conformal window down to $N_f=\frac{3}{2}{N}_c$. $■$

Connections Master

Asymptotic freedom and the running gauge coupling 12.18.03 is the direct upstream input: the electric theory of this unit is asymptotically free exactly when $N_f<3N_c$, the one-loop coefficient $b=3N_c-N_f$ here is the same $b$ that governs the logarithmic running there, and the NSVZ beta function of the Master section is the all-orders supersymmetric completion of that one-loop result. The strong-weak character of Seiberg duality is meaningful only relative to this running: the magnetic theory is weakly coupled precisely in the regime where the electric running has driven $g$ strong.

Theta-vacua and the vacuum angle 12.18.04 supply the second half of the holomorphic scale. The combination ${\Lambda }^b={\mu }^b{e}^{2\pi i\tau }$ with $\tau =\theta /2\pi +4\pi i/g^2$ packages the vacuum angle of 12.18.04 with the coupling of 12.18.03 into the single holomorphic spurion whose broken-$U(1)$ charge drives the entire ADS and duality analysis; without the $\theta$-dependence the selection rules that fix $W_{\mathrm{ADS}}$ would be incomplete, since the anomalous $U(1{)}_A$ rotation acts precisely by shifting $\theta$.

Super-Yang-Mills and the non-renormalization theorem 12.19.04 is the co-produced sibling and the conceptual foundation. The non-renormalization theorem proved there — that the superpotential receives no perturbative corrections — is what licenses the holomorphy argument of this unit; the vector superfield, the field strength $W_{\alpha }$, and the gaugino whose condensate $⟨\lambda \lambda ⟩={\Lambda }^3$ sets the coefficient of $W_{\mathrm{ADS}}$ are all constructed there. This unit is the non-perturbative payoff of that perturbative theorem: the same holomorphy that forbids perturbative corrections also fixes the instanton- and condensate-generated terms.

Spontaneous supersymmetry breaking and the Witten index 12.19.05 is the adjacent application: the Witten index $\mathrm{Tr}(-1{)}^F$ of pure $SU(N_c)$ super-Yang-Mills equals $N_c\ne 0$, which is exactly the statement that SQCD with $N_f<N_c$ does not break supersymmetry but runs away, and the ADS analysis here is the explicit superpotential realisation of that index obstruction.

Seiberg-Witten invariants of four-manifolds 03.07.18 is the deliberate dis-connection. The monopole equations there produce smooth-structure invariants of four-manifolds from the twisted $N=2$ theory; the duality here is an infrared statement about the untwisted $N=1$ gauge theory. They share Seiberg's name and a common origin in supersymmetric gauge dynamics, but one is a topological invariant and the other a dynamical phase diagram — a connection of intellectual lineage, not of curriculum dependency.

Historical & philosophical context Master

The dynamically generated superpotential and the runaway vacuum of $N_f<N_c$ SQCD were found by Ian Affleck, Michael Dine, and Nathan Seiberg in 1984 (Dynamical supersymmetry breaking in supersymmetric QCD, Nucl. Phys. B241, 493–534) ^{[Affleck 1984]}, the first systematic use of holomorphy and instanton calculus to read strong-coupling vacuum structure off weak-coupling data. The exact all-orders beta function that makes the conformal window a precise statement is older still, the Novikov-Shifman-Vainshtein-Zakharov result of 1983 (Exact Gell-Mann-Low function of supersymmetric Yang-Mills theories, Nucl. Phys. B229, 381–393) ^{[NSVZ 1983]}, derived from the anomaly of the supercurrent multiplet. The consistency tool that certifies a proposed infrared description, anomaly matching, is Gerard 't Hooft's 1980 naturalness criterion (Naturalness, chiral symmetry, and spontaneous chiral symmetry breaking) ^{['t Hooft 1980]}, which predates the duality by fourteen years and was repurposed as its sharpest non-dynamical check.

The electric-magnetic duality itself is Seiberg's 1994 result (Electric-magnetic duality in supersymmetric non-Abelian gauge theories, Nucl. Phys. B435, 129–146) ^{[Seiberg 1994]}, with the textbook synthesis of the full phase diagram given by Intriligator and Seiberg in 1995 ^{[Intriligator Seiberg 1995]}. Seiberg duality reaches back to the Montonen-Olive conjecture of the 1970s, that a gauge theory and a theory of its magnetic monopoles describe the same physics with coupling inverted; Montonen-Olive holds exactly in $N=4$ super-Yang-Mills, is realised on Coulomb branches in the $N=2$ Seiberg-Witten solution of the same year, and appears in $N=1$ as the duality of this unit, where the magnetic gauge group is genuinely different from the electric one. The identification of a confined composite operator with an elementary field of a dual gauge theory has no counterpart in non-supersymmetric gauge theory and remains one of the principal pieces of evidence that strongly coupled gauge dynamics admits multiple Lagrangian descriptions.

Bibliography Master

@article{seiberg1994duality,
  author  = {Seiberg, Nathan},
  title   = {Electric-magnetic duality in supersymmetric non-{A}belian gauge theories},
  journal = {Nuclear Physics B},
  volume  = {435},
  pages   = {129--146},
  year    = {1995},
  note    = {hep-th/9411149}
}

@article{ads1984,
  author  = {Affleck, Ian and Dine, Michael and Seiberg, Nathan},
  title   = {Dynamical supersymmetry breaking in supersymmetric {QCD}},
  journal = {Nuclear Physics B},
  volume  = {241},
  pages   = {493--534},
  year    = {1984}
}

@article{nsvz1983,
  author  = {Novikov, V. A. and Shifman, M. A. and Vainshtein, A. I. and Zakharov, V. I.},
  title   = {Exact {G}ell-{M}ann-{L}ow function of supersymmetric {Y}ang-{M}ills theories from instanton calculus},
  journal = {Nuclear Physics B},
  volume  = {229},
  pages   = {381--393},
  year    = {1983}
}

@incollection{thooft1980anomaly,
  author    = {'t Hooft, Gerard},
  title     = {Naturalness, chiral symmetry, and spontaneous chiral symmetry breaking},
  booktitle = {Recent Developments in Gauge Theories},
  editor    = {'t Hooft, G. and others},
  publisher = {Plenum Press, New York},
  pages     = {135--157},
  year      = {1980}
}

@article{intriligatorseiberg1995,
  author  = {Intriligator, Kenneth and Seiberg, Nathan},
  title   = {Lectures on supersymmetric gauge theories and electric-magnetic duality},
  journal = {Nuclear Physics B Proceedings Supplements},
  volume  = {45BC},
  pages   = {1--28},
  year    = {1996},
  note    = {hep-th/9509066}
}

@book{weinberg2000qft3,
  author    = {Weinberg, Steven},
  title     = {The Quantum Theory of Fields, Volume 3: Supersymmetry},
  publisher = {Cambridge University Press},
  year      = {2000}
}