Seiberg–Witten Curves, Differentials, and Periods

A Seiberg–Witten geometry consists of a family of curves, an integral symplectic cycle basis, and a meromorphic differential whose periods are the special coordinates. The curve locates degenerations; the differential fixes dimensions, masses, residues, and normalization. Neither object alone determines the low-energy theory.

Required background. Prepotentials and special Kähler geometry supplies the period relations, and singularities and monodromies supplies the global checks. Helpful background. Branches and analytic continuation explains cut and path conventions.

The geometric data

Many Lagrangian rank-$r$ families admit a genus-$r$ curve presentation. In that case let

$$\pi:\Sigma\longrightarrow\mathcal B^\circ$$

be the family over the regular Coulomb branch and choose cycles

$$A^I,B_I\in H_1(\Sigma_u,\mathbb Z)$$

with intersections

$$A^I\circ A^J=B_I\circ B_J=0, \qquad A^I\circ B_J=\delta^I{}_J.$$

For a meromorphic differential $\lambda_{\mathrm{SW}}$, define

$$a^I(u)=\oint_{A^I}\lambda_{\mathrm{SW}}, \qquad a_{D,I}(u)=\oint_{B_I}\lambda_{\mathrm{SW}}.$$

The normalization is part of $\lambda_{\mathrm{SW}}$; no extra $2\pi$ is implied here. A different convention may put $1/(2\pi i)$ outside every period.

The central charge becomes the period of the cycle representing the electromagnetic charge:

$$Z_\gamma=p^Ia_{D,I}+q_Ia^I+s_am^a.$$

For a principally polarized system, homology intersection supplies the unimodular coordinate pairing, up to the declared orientation. A physical particle lattice can inherit a non-principal multiple or sublattice. In the pure-$SU(2)$ root-unit convention used below, $A\circ B=+1$ and $\delta_{(p,q)}=pB+qA$, so

$$\delta_\gamma\circ\delta_{\gamma'} =-(pq'-qp'), \qquad \langle\gamma,\gamma'\rangle_{\mathrm{part}} =-2\,\delta_\gamma\circ\delta_{\gamma'}.$$

More general Seiberg–Witten systems are naturally described by a polarized abelian variety or Prym of complex dimension $r$; an auxiliary spectral curve can have genus larger than the Coulomb rank. Genus $r$ is therefore a property of the hyperelliptic families under discussion, not a universal definition.

The defining differential condition

The essential relation is that derivatives of $\lambda_{\mathrm{SW}}$ with respect to Coulomb moduli are holomorphic one-forms up to exact terms:

$$\frac{\partial\lambda_{\mathrm{SW}}}{\partial u^k} =c_{kI}(u)\,\omega^I+d_xf_k.$$

Exact terms have zero periods on closed cycles. Therefore

$$\frac{\partial a^I}{\partial u^k} =c_{kJ}\oint_{A^I}\omega^J, \qquad \frac{\partial a_{D,I}}{\partial u^k} =c_{kJ}\oint_{B_I}\omega^J.$$

The effective coupling is

$$\tau =\frac{\partial a_D}{\partial u} \left(\frac{\partial a}{\partial u}\right)^{-1}.$$

It is the Riemann period matrix in the corresponding normalized basis, hence symmetric with positive imaginary part on a regular physical patch.

Ambiguities and physical normalization

Adding an exact differential

$$\lambda_{\mathrm{SW}}\longmapsto \lambda_{\mathrm{SW}}+d_xf$$

does not change closed periods, provided $f$ is single-valued along the cycles and no endpoint or residue contribution is introduced. Adding a holomorphic differential does change periods and generally corresponds to a redefinition that must be matched to ultraviolet asymptotics.

With massive matter, $\lambda_{\mathrm{SW}}$ has poles whose residues are linear combinations of flavor masses. Moving a pole across a cycle shifts an electromagnetic period by a mass, reflecting the affine flavor extension of the charge system. Residues, paths, and flavor-charge normalization must be specified together.

The overall scale is fixed by a physical anchor, commonly

$$a\sim\text{classical adjoint expectation value}$$

at weak coupling and by the known one-loop logarithm in $a_D$. Choosing the normalization after inspecting desired monodromies is circular.

Hyperelliptic curves and branch cuts

For a hyperelliptic presentation

$$y^2=P_{2r+2}(x;u,m,\Lambda),$$

the curve is a two-sheeted cover of the $x$-plane branched at roots $e_i$. A standard cut system pairs branch points, and cycles encircle or connect cuts. The visual placement of cuts is conventional; their homology intersections and continuation are physical inputs.

When two branch points collide, a cycle can shrink. The polynomial discriminant

$$\operatorname{Disc}_xP \propto\prod_{i<j}(e_i-e_j)^2$$

vanishes. Determining which cycle vanishes requires the cut system, not only the discriminant value.

As $u$ moves, branch points braid. Continuously transporting both square-root sheets and cycles produces an integral monodromy. Re-sorting roots numerically at every step without continuity can jump between sheets and return a false matrix.

Pure SU(2) normalization and exact fixture

Use the chapter benchmark convention, with $\Lambda>0$, $u_b=2\Lambda^2$, and

$$y^2=(x-u)(x^2-\Lambda^4)$$

and

$$\lambda_{\mathrm{SW}} =\frac{\sqrt2}{2\pi} \frac{(x-u)\,dx}{y}.$$

Direct differentiation gives, up to the displayed representative with no further exact term,

$$\frac{\partial\lambda_{\mathrm{SW}}}{\partial u} =-\frac{\sqrt2}{4\pi}\frac{dx}{y},$$

the unique holomorphic differential on the genus-one curve. The factor is chosen so that, with the corresponding $A$ cycle,

$$a(u)\sim\sqrt{2u}$$

for large positive $u$. The $B$ period then reproduces the one-loop logarithm.

At the base fiber, take cuts $[-\Lambda^2,+\Lambda^2]$ and $[u_b,\infty]$. Orient $A$ around the first cut and $B$ around the lifted $+\Lambda^2$-to-$u_b$ contour so that

$$A\circ B=+1, \qquad a(u_b)>0, \qquad a_D(u_b)\in i\mathbb R_{>0}.$$

The branch points are

$$e_1=-\Lambda^2, \qquad e_2=+\Lambda^2, \qquad e_3=u,$$

plus the point at infinity. The exact polynomial discriminant is

$$\operatorname{Disc}_x \bigl[(x-u)(x^2-\Lambda^4)\bigr] =4\Lambda^4(u^2-\Lambda^4)^2.$$

Collisions at $u=\pm\Lambda^2$ give the two finite singular fibers. The based paths in the chapter convention call the positive point the monopole point and the negative point the dyon point. The curve, differential, and its two singularities are fixed in Seiberg and Witten 1994, §§5–6.

For real $u>\Lambda^2$, set

$$s=u+\Lambda^2, \qquad m=\frac{2\Lambda^2}{s}.$$

With $K(m)$ and $E(m)$ denoting complete elliptic integrals in the parameter convention,

$$\begin{aligned} a(u) &=\frac{2\sqrt{2s}}{\pi}E(m),\\ a_D(u) &=\frac{2i\sqrt{2s}}{\pi} \bigl[K(1-m)-E(1-m)\bigr],\\ \tau(u) &=i\,\frac{K(1-m)}{K(m)}. \end{aligned}$$

These formulas fix the sheet and cycle orientations rather than merely asserting that an elliptic representation exists. At the frozen base point,

$$\begin{aligned} \frac{a(u_b)}{\Lambda} &=1.966685301550330\ldots,\\ \frac{a_D(u_b)}{i\Lambda} &=0.473434436535537\ldots,\\ \frac{\tau(u_b)}{i} &=0.854584443278744\ldots. \end{aligned}$$

The deterministic structured period-map record evaluates the integrals independently and fixes its numerical tolerance before comparison. The point $u_b$ is in the weak-coupling chamber; this fixture tests periods and transport but does not assert a complete chamber spectrum.

Periods and Picard–Fuchs equations

Instead of integrating cycles separately at every $u$, periods often satisfy a differential equation. For rank one, one can derive a second-order Picard–Fuchs equation by differentiating $\lambda_{\mathrm{SW}}$ and reducing meromorphic forms modulo exact terms:

$$\mathcal L_u\,\Pi(u)=0.$$

Two independent solutions give $a$ and $a_D$. Boundary conditions at weak coupling fix their linear combination. Analytic continuation of the solutions produces the same monodromy as transporting cycles. Higher-rank hyperelliptic curves and their semiclassical monodromy constraints are proposed and checked in Klemm, Lerche, Theisen, and Yankielowicz 1995, pp. 5–8, Eqs. (15)–(20); that paper does not supply every detailed period and finite-locus monodromy calculation.

The differential equation is efficient but does not remove branch choices. Its singular points, Frobenius exponents, logarithms, and connection matrices must be tied back to an integral cycle basis.

Numerical period evaluation

A reproducible quadrature should record:

1. numerical values and precision for $u,m,\Lambda$;
2. labeled branch points before and after continuation;
3. explicit contours and sheet choices;
4. endpoint regularization or a variable change for square-root singularities;
5. at least two precisions and contour resolutions;
6. weak-coupling or local-series benchmarks;
7. an integer-recognition tolerance fixed before extracting monodromy.

To continue around a discriminant point, move $u$ in small steps and continue roots and cycles by proximity plus topology. Evaluate both periods, fit the final vector to $M\Pi$, and require the same integral $M$ at increasing precision. Then transform charges contragrediently and verify $Z$.

What the curve does not decide alone

Several inequivalent physical theories can share an algebraic curve while differing in:

• the differential normalization or mass residues;
• the integral sublattice of genuine lines;
• the global form and discrete theta angle;
• the BPS chamber and populated charges;
• boundary or defect data.

The complete Seiberg–Witten solution includes all of these layers relevant to the claimed observables.

Common pitfalls

Quoting a curve without a differential. The curve gives degeneration topology but not physical period normalization or flavor residues.

Sorting branch points independently at each parameter value. This destroys analytic continuation and can fabricate monodromy.

Recognizing a near-integer matrix at one precision. Numerical monodromy needs convergence, an error budget, and an independent asymptotic check.

Exercises

For the pure-$SU(2)$ curve:

1. compute the collision loci and exact polynomial discriminant;
2. explain which cycle and path data identify the massless charge;
3. evaluate the three dimensionless base-point quantities above from the elliptic-integral formulas and check $\operatorname{Im}\tau>0$.

Solution

The three root differences give

$$\operatorname{Disc}_x =4\Lambda^4(u-\Lambda^2)^2(u+\Lambda^2)^2,$$

so the finite singularities are $u=\pm\Lambda^2$. The discriminant identifies these values but not the homology class of the shrinking loop. One must specify branch cuts, the $A/B$ cycle basis, and a path from each singularity to the base point. The resulting vanishing cycle gives the electromagnetic charge and its pairing after the polarization crosswalk.

At $u_b=2\Lambda^2$, $m=2/3$. Direct numerical evaluation gives the three quoted values. Since both $K(1/3)$ and $K(2/3)$ are positive real numbers,

$$\operatorname{Im}\tau(u_b) =\frac{K(1/3)}{K(2/3)}>0.$$

References

• Klemm, Albrecht, Wolfgang Lerche, Stefan Theisen, and Stefan Yankielowicz. “Simple Singularities and $\mathcal N=2$ Supersymmetric Yang–Mills Theory.” Physics Letters B 344 (1995): 169–175. arXiv:hep-th/9411048.
• Seiberg, Nathan, and Edward Witten. “Electric–Magnetic Duality, Monopole Condensation, and Confinement in $\mathcal N=2$ Supersymmetric Yang–Mills Theory.” Nuclear Physics B 426 (1994): 19–52; erratum 430 (1994): 485–486. arXiv:hep-th/9407087.