Reading the second Seiberg–Witten paper

Why does the same pure SU(2) theory appear with two different elliptic curves? This question makes a useful entrance to Nathan Seiberg and Edward Witten’s Monopoles, Duality and Chiral Symmetry Breaking in N=2 Supersymmetric QCD, hep-th/9408099v1, 17 August 1994. We use that preprint, whose cover identifies v1; the PDF links below count its cover as page 1. Its journal publication is Nuclear Physics B 431 (1994), 484–550.

This guide selects the one-flavor example and the adjoint-hypermultiplet deformation. Read fundamental matter and decoupling first. The later adjoint-mass section additionally uses N=4 and N=2-star. These examples have global gauge group SU(2), rank one, in flat four-dimensional spacetime. A full hypermultiplet mass preserves N=2, with eight real Poincaré supercharges and four manifest in N=1 notation. Only the zero-mass adjoint theory enhances to sixteen.

Read the convention note before a matter formula

The introduction first recalls the original pure-theory formulas. Its subsequent convention note, (1.6), changes the electric unit for fundamental matter. To keep that change visible, call the second-paper fundamental periods $A,B$, although the paper prints $a,a_D$.

Quantity	First-paper units used on this site	Second-paper fundamental units
Electric and magnetic periods	$a,\ a_D$	$A=a/2,\ B=a_D$
Electric and magnetic labels	$n_e,\ n_m$	$q=2n_e,\ p=n_m$
W boson; fundamental electric label	$1;\ \pm1/2$	$2;\ \pm1$
Derivative coupling	$\tau_s=da_D/da$	$\tau_f=dB/dA=2\tau_s$
Low-energy coupling convention	$\theta_{\rm eff}/(2\pi)+4\pi i/g_{\rm eff}^2$	$\theta_{\rm eff}/\pi+8\pi i/g_{\rm eff}^2$
Modulus and complex masses	$u=\langle\operatorname{Tr}\phi_{\rm SW}^2\rangle,\ m_i$	The same $u,m_i$
Pure curve scale	Singularities at $u=\pm\Lambda_0^2$	The same singularity calibration
One-flavor scale	A different theory needs its own $\Lambda_1$	$\Lambda_0^4=m\Lambda_1^3$ in the curve’s scheme

Here $\phi_{\rm SW}=g\phi_{\rm can}$ fixes the scalar normalization, and the trace is the ordinary fundamental trace. Changing period units does not divide a physical mass or $u$ by two. The coupling row compares two descriptions at the same low-energy point; neither derivative coupling should be replaced by a microscopic running coupling throughout the Coulomb branch. Nor is $a_D/a$ generally the derivative ratio.

For a charge with flavor weights $S_i$, the invariant comparison is

$$Z=n_m a_D+n_e a+\sum_i\frac{S_i m_i}{\sqrt2} =pB+qA+\sum_i\frac{S_i m_i}{\sqrt2}, \qquad M_{\rm BPS}=\sqrt2\,|Z|. \tag{PC11.1}$$

Thus the elementary charges $p=0,q=\pm1,S=1$ give $M=|m\pm\sqrt2 A|$. At weak coupling $A\simeq\sqrt{2u}/2$, the vanishing condition is $u\simeq m^2$, not $m^2/4$. This is a quick way to detect use of the adjoint charge in a fundamental formula. The canonical mass calculation derives the mass from the superpotential; source (5.3) supplies the exact central charge. A permitted charge does not by itself establish a stable BPS particle.

A selected reading order

Use each row to answer its question before moving on. Printed page numbers exclude the cover.

Paper passage	Question to answer	Canonical calculation
Convention note, (1.6), pp. 5–6	Which period changes when the elementary electric label becomes an integer?	Charge dictionary
(3.1), p. 10; (5.3), p. 19	Does the central charge reproduce the actual fundamental mass?	Matter mass
(6.1)–(6.4), p. 20; (15.1)–(15.4), pp. 51–52	Why can massive periods acquire additive flavor shifts?	Differential and residues
(11.1)–(11.7), pp. 37–38; (12.5)–(12.8), p. 45	Why does removing a fundamental produce a different pure curve?	Heavy-flavor limit and the reconstruction below
(16.1)–(16.3), p. 56; (16.10), p. 58	Which normalization is used for the adjoint theory?	Adjoint periods and coupling
(16.17)–(16.25), pp. 61–63	Which modulus and scale stay fixed in the adjoint decoupling limit?	Two different limits

The residue passage is important even when only period derivatives are needed for a metric. A cycle moved past a pole changes its integral by $2\pi i$ times the residue. For massive matter this can be proportional to $m_i/\sqrt2$; compensating flavor-charge coordinates keep (PC11.1) fixed. The elementary-threshold exercise works out the affine monodromy. Pure two-by-two monodromies alone do not contain that mass information.

Restore the map between the two pure curves

The source’s (11.7) states the degree-two relation, called an isogeny: a map of smooth elliptic curves which changes the integral period lattice by a finite index. Let $s=\Lambda_0^2\ne0$, take $u\ne\pm s$, and distinguish the two sets of coordinates:

$$\begin{aligned} E_f:\quad &y^2=x\left(x^2-ux+\frac{s^2}{4}\right),\\ E_s:\quad &Y^2=(X^2-s^2)(X-u). \end{aligned} \tag{PC11.2}$$

Both have $[x]=[X]=[u]=[s]=2$ and $[y]=[Y]=3$. Here is an explicit reconstruction of the map:

$$X=x+\frac{s^2}{4x},\qquad Y=-y\left(1-\frac{s^2}{4x^2}\right). \tag{PC11.3}$$

It is enough first to work away from $x=0$. Put $b=s^2/4$. Then $X-u=(x^2-ux+b)/x$ and $X^2-s^2=(x^2-b)^2/x^2$. Their product is $y^2(1-b/x^2)^2=Y^2$. Conversely, a generic $X$ has two preimages satisfying $x^2-Xx+b=0$. This is a degree-two map, not an invertible renaming of a coordinate. The apparent affine singularities extend to the compact smooth curves; for example both $x=0$ and the point at infinity map to the target point at infinity.

The differential matters as much as the cubic. In the established sheet conventions, use

$$\lambda_s=\frac{\sqrt2}{2\pi}\frac{(X-u)\,dX}{Y}, \qquad \lambda_f=-\frac{\sqrt2}{4\pi}\frac{y\,dx}{x^2}. \tag{PC11.4}$$

The first is the original differential recalled in source (1.4); the second is the pure limit of the one-flavor differential. It obeys $\partial_u\lambda_f=\sqrt2\,dx/(8\pi y)$, matching source (15.2). Since $dX/Y=-dx/y$, substitution gives the exact identity

$$\varphi^*\lambda_s =-\frac{\sqrt2}{2\pi}\frac{x^2-ux+b}{xy}\,dx =-\frac{\sqrt2}{2\pi}\frac{y}{x^2}\,dx =2\lambda_f. \tag{PC11.5}$$

The symbol $\varphi^*$ means “substitute the map into the differential.” It follows that periods must be compared with the images of their cycles. Choose the real weak branch $u>s>0$, positive electric periods, and magnetic periods positive imaginary and vanishing at $u=s$. The image of the fundamental electric cycle is the site electric cycle; the image of its magnetic cycle is twice the site magnetic cycle. To see the doubling, let $e_-<e_+$ be the nonzero roots of $E_f$. Both endpoints map to $X=u$, while the intermediate point $x=s/2$ maps to $X=s$. One traversal of the interval between $e_-$ and $e_+$ therefore goes from $u$ to $s$ and back, changing target sheet at $s$. Returning on the other source sheet traverses that target cycle again. The electric cycle maps to the complementary primitive cycle, with its sign fixed by the positive electric period. Consequently $a=2A$ and $a_D=B$. The image cycle lattice has index two. Treating the map as an integral determinant-one basis change would lose precisely this factor.

We can check both cycle statements without relying on the appearance of the cubics. Let $e_\pm=(u\pm\sqrt{u^2-s^2})/2$ be the nonzero roots of $E_f$ and define $k_f=e_-/e_+$, $k_s=2s/(u+s)$. With $K(k)=\int_0^{\pi/2}(1-k\sin^2t)^{-1/2}dt$—the argument is the parameter, not its square root—the derivative periods are

$$\begin{aligned} A'&=\frac{\sqrt2}{2\pi\sqrt{e_+}}K(k_f),& B'&=\frac{i\sqrt2}{2\pi\sqrt{e_+}}K(1-k_f),\\ a'&=\frac{\sqrt2}{\pi\sqrt{u+s}}K(k_s),& a_D'&=\frac{i\sqrt2}{\pi\sqrt{u+s}}K(1-k_s). \end{aligned} \tag{PC11.6}$$

Primes mean $u$ derivatives. The formulas follow by replacing the coordinate on each branch cut by its endpoint plus a squared sine; the two factors that vanish at endpoints then cancel against $dx$. They give $a'=2A'$ and $a_D'=B'$, and hence the opening coupling dictionary.

As a check on the absolute periods as well, take $s=1,u=3$. Integrating $\lambda_f$ on closed contours around its two adjacent cuts, with a continuous square-root sheet, gives

Period	Numerical value
$A$	$1.2160067234$
$B/i$	$0.9064939198$
$a$	$2.4320134468$
$a_D/i$	$0.9064939198$

The site values can instead be obtained from the ordinary convergent real integrals

$$a=\frac{\sqrt2}{\pi}\int_0^\pi\sqrt{u-s\cos t}\,dt, \qquad a_D=\frac{i\sqrt2}{\pi}\int_s^u \sqrt{\frac{u-X}{X^2-s^2}}\,dX. \tag{PC11.7}$$

One should not integrate $\lambda_f$ straight through its pole at $x=0$. A closed contour avoiding the pole is well defined. Checking only (PC11.6) would also leave additive constants in the periods undetected. These absolute-period checks fix that remaining ambiguity in the stated pure-theory branch.

Follow one heavy fundamental through the paper

Read (12.5) together with the sentence fixing its scheme. In the source’s scale calibration,

$$\Lambda_0^4=m\Lambda_1^3,\qquad y^2=x^2(x-u)+\frac{m\Lambda_1^3x}{4} -\frac{\Lambda_1^6}{64}. \tag{PC11.8}$$

At fixed $\Lambda_0$ and finite $u$, the last term becomes $-\Lambda_0^8/(64m^2)$ and vanishes as $|m|\to\infty$; the linear term remains. The result is $E_f$, not $E_s$ with the same coordinates. The coefficient $-1/64$ is part of the quantum curve input; the quark degeneration check tests its physical mass normalization. Finite changes of scale scheme can change a matching coefficient. A complex mass phase cannot be discarded as such a numerical convention.

The low-energy comparison requires energies and Coulomb masses small relative to $|m|$ and avoidance of an additional light threshold. It does not say that the flavored periods equal the pure periods plus a mass term at finite $m$.

Return to the adjoint units in section 16

For this section, first read the adjoint mass lesson. In (16.1)–(16.3), the paper explicitly uses the original electric period for the adjoint theory: $a\simeq\sqrt{2u}$, $\tau=\theta/(2\pi)+4\pi i/g^2$, and a derivative differential with coefficient $\sqrt2/(4\pi)$. Its fundamental-matter coefficient is $\sqrt2/(8\pi)$. The massless adjoint theory has $a_D=\tau a$ exactly; this special linear relation must not be used for a general massive theory.

The mass $m$ belongs to the full adjoint hypermultiplet. Its charged masses are $|m\pm\sqrt2 a_{\rm cl}|$, and its neutral mass is $|m|$. It is not a mass for the chiral inside the N=2 vector multiplet. Thus its nonzero value leaves eight charges, while $m=0$ restores sixteen and a neutral hypermultiplet on a generic Coulomb branch. Keeping only a vector prepotential does not remove that massless neutral field.

Use $Q=e^{2\pi i\tau}$ for the adjoint instanton parameter, reserving $q$ above for an electric charge. Equations (16.18), (16.23)–(16.25) relate the modular coordinate to the physical weak-cusp prescription for $u$:

$$\widetilde u=u-\frac{e_1m^2}{8},\qquad f=\frac{m^2}{4},\qquad e_1=\frac23+16Q+O(Q^2). \tag{PC11.9}$$

The coefficient $e_1$ is defined by the theta series in the curve lesson; the displayed expansion is source (16.14). Its theta labels differ from modern labels, as that lesson specifies. In the shifted cubic (16.19), the three roots approach $u,+\Lambda_0^2,-\Lambda_0^2$ under

$$|m|\to\infty,\quad Q\to0,\quad \Lambda_0^4=4m^4Q\ \text{fixed},\quad u\ \text{fixed}. \tag{PC11.10}$$

The result is directly $E_s$, as (16.22) states. This is a different threshold law from a single fundamental. In particular, the scale is fixed by $2m^2\sqrt Q=\Lambda_0^2$ on a continuously chosen phase branch. Keeping $\widetilde u$ fixed instead selects a different region; the last exercise diagnoses it.

Taking $m\to0$ at fixed $\tau$ is the other limit: it restores N=4, with no generated pure-theory scale held fixed. These are statements about the flat-space physical mass. An equivariant calculation can use a shifted mass parameter; the regulator translation must be applied separately. The modular behavior of a curve also does not by itself specify the global form and spectrum of line operators of a quantum gauge theory.

Exit calculations

1. Change the period units in a monodromy

In the pure site basis $\Pi_s=(a_D,a)^T$, take $M_+=\left(\begin{smallmatrix}1&0\\-2&1\end{smallmatrix}\right)$ and $M_\infty=\left(\begin{smallmatrix}-1&2\\0&-1\end{smallmatrix}\right)$. Set $\Pi_f=(B,A)^T=D\Pi_s$, where $D=\operatorname{diag}(1,1/2)$. Find both new monodromies and the charge-row transformation for $M_+$. Does the fact that the answers have integer entries make $D$ a duality transformation within the original integral basis?

Hint 1

Continue the periods and then change units. Compare that result with changing units first.

Hint 2

Use $M_f=DM_sD^{-1}$ and require $(p',q')M_f=(p,q)$.

Solution

The results are

$$M_{+,f}=\begin{pmatrix}1&0\\-1&1\end{pmatrix}, \qquad M_{\infty,f}=\begin{pmatrix}-1&4\\0&-1\end{pmatrix}.$$

For the first loop, $B'=B$, $A'=A-B$, and $(p',q')=(p+q,q)$. Hence $(p+q)B+q(A-B)=pB+qA$. The magnetic period still vanishes at the monopole point, while the electric period picks up one magnetic period in the new units. These are the matrices in source (11.5)–(11.6).

But $\det D=1/2$, not one. The physical antisymmetric pairing is $qp'-pq'=2(n_en_m'-n_mn_e')$ between two charge vectors; the primes in this pairing label the second vector. The rescaling changes the coordinates used for the lattice and must be accompanied by this dictionary. Integer entries of these particular conjugated monodromies do not make $D$ an integral change of basis.

2. Wind the heavy mass phase

At a large fixed $|m|$, continue $m=m_0e^{i\varphi}$ from $\varphi=0$ to $2\pi$. Compare the fundamental law $\Lambda_0^4=m\Lambda_1^3$ at fixed $\Lambda_1^3$ with the adjoint law $\Lambda_0^4=4m^4Q$ at fixed $Q$. How often does $\Lambda_0^4$ wind, and what happens to continuously labeled roots $u_\pm=\pm\Lambda_0^2$? What must change if the low-energy scale is instead to stay fixed throughout?

Hint 1

Track the argument of $\Lambda_0^4$ continuously before taking its square root.

Hint 2

A square root changes sign after one winding of its argument. The two threshold laws contain different powers of the same complex mass.

Solution

In the fundamental case $\Lambda_0^4$ winds once, so its continuously transported square root changes sign. The two finite pure singularity labels are exchanged; their unordered set returns to itself. In the adjoint case $\Lambda_0^4$ winds four times, and $\Lambda_0^2$ winds twice and returns to its starting value without exchanging the labels.

To keep $\Lambda_0^4$ fixed, change $\Lambda_1^3$ by $e^{-i\varphi}$ in the first case, or $Q$ by $e^{-4i\varphi}$ in the second. These continuations can remain at weak coupling and large mass. They describe the limiting pure curves and their scale branches, not a claim that every BPS state stays stable along the path, nor a count of N=1 vacua. At finite heavy mass there are suppressed curve corrections. Replacing $m$ by $|m|$ would erase the winding altogether.

3. Hold the wrong modulus fixed

In the adjoint limit fix $\Lambda_0^4=4m^4Q$ and let positive $m$ grow. A proposed comparison also fixes $\widetilde u$, rather than $u$. Use (PC11.9) through its first correction to find $u(m)$. Does this approach the finite-$u$ pure theory? Specify the $\widetilde u(m)$ needed for that intended comparison.

Hint 1

Substitute $Q=\Lambda_0^4/(4m^4)$ into the shift before taking the limit.

Hint 2

Compare the resulting scale of $u$ with $a_{\rm cl}^2/2=u_{\rm cl}$ at the weak cusp.

Solution

The shift gives

$$u=\widetilde u+\frac{m^2}{12} +\frac{\Lambda_0^4}{2m^2} +O\!\left(\frac{\Lambda_0^8}{m^6}\right).$$

Thus fixed $\widetilde u$ sends $u$ to infinity. At leading weak coupling $a/m\to1/\sqrt6$, rather than zero. It fails the fixed finite Coulomb-scale comparison; this does not make it an impossible family of vacua, only a different scaling region.

For fixed $u=u_*$ one must choose $\widetilde u=u_*-m^2/12-\Lambda_0^4/(2m^2) +O(\Lambda_0^8/m^6)$. Then the large coordinate shift cancels and the two finite roots approach $\pm\Lambda_0^2$. The massless limit $m\to0$ at fixed coupling cannot substitute for this limit: it restores the adjoint hypermultiplet instead of integrating it out.

After these checks, return to (12.8) and (16.22) side by side. You should be able to explain their different cubics, the unchanged physical mass, and the different data held fixed in their limits. A next focused reading is the residue argument following (16.22), which fixes $f=m^2/4$ beyond its weak-coupling match; its canonical normalization check provides the entry.