Calabi–Yau integrals

The current top of the function-class ladder: integrals whose geometry is a Calabi–Yau threefold, where no ready-made dictionary of functions exists and the answer must be built from integrals over the geometry itself.

The content on this page was written by AI under human supervision.

The ladder: The BootLoops integral class · Polylogarithms · Elliptic · K3 · Calabi–Yau

The top of the ladder so far

Every rung of this ladder has been a statement about the geometry an integral carries — read, for a Feynman graph, from its maximal cutthe integral with every internal propagator put on shell; it solves the homogeneous part of the diagram's differential equation and carries the diagram's geometry. At the bottom that geometry was a sphere and the answer was polylogarithmic. One rung up it was an elliptic curve; one rung above that, a K3 surface. Those last two objects were introduced separately, but they are the first two members of a single family, the Calabi–Yau manifolds.

A Calabi–Yau $n$-fold is a compact complex manifold of complex dimension $n$ that carries a nowhere-vanishing holomorphic $n$-form. Take the three ingredients in turn. A complex manifold of dimension $n$ is a space covered by coordinate charts to $\mathbb{C}^n$ whose transition functions are holomorphic — the complex-analysis notion of smooth, applied to the glue between charts. Compact means closed and without boundary, like the torus and unlike the plane: integrals over the whole space converge because there is no infinity to run off to. And a holomorphic $n$-form is an object that looks locally like $f(z_1,\dots,z_n)\,dz_1\wedge\cdots\wedge dz_n$ with $f$ holomorphic, the $\wedge$ being the antisymmetric product that makes the differentials transform correctly under changes of variables; "nowhere vanishing" demands an $f$ with no zeros in any chart. It is the thing a complex analyst integrates — the curved-space generalization of the $dz$ in a contour integral — and requiring one to exist globally, with no zeros or poles, is a strong constraint on the shape of the space.

For $n=1$ the definition picks out exactly the torus: on the elliptic curve $y^2 = x^3 + ax + b$ the form is $dx/y$, the same differential whose integrals gave the previous rungs their elliptic functions. So an elliptic curve is a Calabi–Yau one-fold. For $n=2$ the definition (with a simple-connectedness condition that rules out four-tori) picks out the K3 surfaces of the last rung: a K3 is a Calabi–Yau two-fold. The next member, the Calabi–Yau threefold, is where this page lives.

The standard example of a threefold is the quintic. Complex projective space $\mathbb{P}^4$ is the space of complex lines through the origin of $\mathbb{C}^5$, with homogeneous coordinates $[z_1:\cdots:z_5]$. Inside it, take the zero locus of a degree-five homogeneous polynomial, for instance the Fermat quintic

$$z_1^5 + z_2^5 + z_3^5 + z_4^5 + z_5^5 \;=\; 0.$$

This is a compact complex three-dimensional space, and the degree is not negotiable: in $\mathbb{P}^4$, degree five — the number of homogeneous coordinates — is exactly the degree at which the hypersurface admits a nowhere-vanishing holomorphic three-form. Lower degree and the would-be form has poles; higher degree and it has zeros.

Eugenio Calabi conjectured in 1954 that a compact complex manifold of the kind just defined — he phrased the condition as the vanishing of the first Chern class — always admits a metric with zero Ricci curvature, the vacuum Einstein condition. Shing-Tung Yau proved the conjecture in 1977–78, which is why the spaces carry both names. String theory then adopted them: Candelas, Horowitz, Strominger, and Witten argued in 1985 that compactifying the ten-dimensional heterotic string on a Calabi–Yau threefold preserves one four-dimensional supersymmetry, and the spaces became the standard candidate shapes for the six hidden dimensions.

For everything below, the metric side of the story can be set aside. The Feynman-integral rungs use only the complex-analytic definition — the holomorphic $n$-form and the numbers you get by integrating it.

Periods and the fourth-order equation

Integrate the holomorphic $n$-form over a closed $n$-dimensional cycle and you get a number, called a period of the manifold. The prototype lives one complex dimension down: $\oint dz/z = 2\pi i$ is a period, and the two integrals of $dx/y$ over the two independent loops of an elliptic curve are the periods that governed the elliptic rung. A Calabi–Yau threefold has finitely many independent three-cycles, so it has finitely many independent periods.

Now let the threefold vary in a family with one complex parameter $z$ — for a Feynman integral, $z$ will be a kinematic variable like $p^2$. The periods become functions $\varpi(z)$, and they obey a linear ordinary differential equation for a reason that fits in one sentence: differentiating the period integral under the integral sign again and again produces forms that all live in a fixed finite-dimensional cohomologythe space of closed forms modulo exact ones; its dimension counts the independent things a cycle can measure, so integrals of any more forms than that must be linearly related space, so some finite set of derivatives of $\varpi$ must be linearly dependent over functions of $z$. The resulting equation is called the Picard–Fuchs equation of the family. For the families relevant here the third cohomology is four-dimensional, so the equation has order four: four independent periods, one fourth-order operator.

The classic example is the family mirror to the quintic, the case worked out by Candelas, de la Ossa, Green, and Parkes. Its Picard–Fuchs operator, with $\theta = z\,d/dz$, is

$$L \;=\; \theta^4 \;-\; 5z\,(5\theta+1)(5\theta+2)(5\theta+3)(5\theta+4),$$

and the unique power-series solution at $z=0$ is

$$\varpi_0(z) \;=\; \sum_{n\ge 0} \frac{(5n)!}{(n!)^5}\, z^n.$$

Two structural features of such operators matter for the ladder. First, they are self-dual: the four-dimensional solution space carries a constant antisymmetric pairing, built from Wronskian-type combinations, which the differential equation preserves — the differential-equation shadow of the pairing between three-cycles on the threefold. Operators with this structure, integer series coefficients, and the right local behavior are called Calabi–Yau operators, and they have been cataloged by mathematicians in their own right. Second, the point $z=0$ above is a point of maximal unipotent monodromya MUM point: a singular point of the equation where one solution is a power series and the other three pick up one, two, and three powers of $\log z$ — the most degenerate logarithmic structure the order allows, and the standard anchor for Calabi–Yau period expansions, the expansion point around which almost everything known about these periods is organized.

Why anyone outside algebraic geometry cares about the fourth-order equation comes down to one 1991 calculation. Candelas, de la Ossa, Green, and Parkes solved the period equation of the mirror family and read off, from the coefficients of its solutions, predictions for the number of rational curves of each degree on the quintic itself — 2875 lines, 609,250 conics, 317,206,375 twisted cubics, and on up. The first two numbers were known to algebraic geometers — the 2875 lines classically, the conic count from the 1980s — but the degree-three number was not yet known, and when Ellingsrud and Strømme computed it independently their answer initially disagreed; the error turned out to be in their program, and the corrected count matched the physicists' prediction. The period equation had produced counts of curves that geometers had yet to obtain, and the demonstration launched mirror symmetry as a mathematical subject. For our purposes the moral is narrower but just as useful: the fourth-order equation is a rigid, arithmetically special object, and its solutions carry exact information.

Where the integral class meets them

The cleanest route from Feynman diagrams to Calabi–Yau geometry is the banana ladder: the $l$-loop two-point integral with $l+1$ massive propagators strung between the same two vertices. The two-loop banana is the sunrise, whose maximal cut lives on an elliptic curve. The three-loop banana's cut is the period of a K3 surface. The pattern continues: the $l$-loop equal-mass banana is governed by a Calabi–Yau $(l-1)$-fold, one complex dimension per extra loop — an identification developed in the line of work of Bönisch, Duhr, Klemm, Nega, and collaborators. So the four-loop banana is the ladder's first Calabi–Yau threefold rung — the first place a physical two-point function is governed by the same class of geometry that string theory compactifies on. So far the rung belongs to the Feynman branch of the integral class: the lattice branch tops out at the K3 of the Watson problem, and the threefold is where the collider and gravity integrals lead.

The equal-mass four-loop banana makes the meeting concrete. Its maximal cut satisfies an order-four Picard–Fuchs operator known in the mathematicians' catalog as AESZ-34 — the operator was sitting in the Calabi–Yau tables before the physics arrived. On the BootLoops side, the banana page records the full symbolic solution of this rung: the system of five basis integrals solved as an exact first-order system in $\theta$, regular at the MUM point, every series coefficient generated by an integer recursion, the boundary data written exactly in terms of $\pi$ and zeta values, and the whole construction validated against independent evaluations to at least 67 digits.

The site's CY₃ threshold page then pushes past the MUM point to a locus the standard treatments do not cover: a particle-production threshold. Tune the four-loop banana's masses-squared to $(1,1,1,1,16)$, so the heavy line sits exactly on the production threshold of the other four. On that one-parameter slice the operator has order six, and its local exponents at the threshold — the leading powers available to series solutions there — are $\{1,1,2,2,\tfrac54,\tfrac74\}$: two quarter-integer exponents, meaning two of the six periods rotate by $\pm i$ under one walk around the threshold — monodromythe linear map on the solution space induced by carrying the integration contour once around a singular point and back of order four, something no expansion of logarithms around the MUM point can produce. The two connection coefficients that project the holomorphic period onto the fractional branches have exact values built from the lemniscatic constant $\Gamma(\tfrac14)$ and powers of $\pi$:

$$c_{5/4}\;=\;-\frac{1}{\sqrt{2\pi}\,\Gamma(\tfrac14)^2},\qquad c_{7/4}\;=\;-\frac{5\,\Gamma(\tfrac14)^2}{384\sqrt2\,\pi^{5/2}},\qquad c_{5/4}\,c_{7/4}\;=\;\frac{5}{768\,\pi^3}\ \ \text{exactly},$$

validated against independent evaluations to at least 71 digits. So far as we could determine, these are the first exact $\Gamma(\tfrac14)$ constants obtained anywhere in the banana ladder — one rung below, at the K3 threshold, the analogous constant was $-\sqrt3/(36\pi)$, and the constant's ingredients change because the geometry does.

The same geometry has reached gravity. In the post-Minkowskian expansion of black-hole scattering — the calculation behind gravitational-wave predictions — the integrals climb the identical ladder as the order in Newton's constant grows: polylogarithms through third order, a first K3 surface at fourth, and at fifth order a complete classification finds exactly two Calabi–Yau threefolds and two K3 surfaces across the order's integral families. At fifth order and first order in the mass ratio, the radiated energy is built from the periods of a Calabi–Yau threefold. The black-hole scattering page carries that story, including the first evaluations of these K3 and Calabi–Yau periods on bound orbits, the case relevant to gravitational-wave observations.

The working method itself changes at this rung. On the polylogarithmic rungs one expands in a known basis of functions and fits. Here the function space is built directly from the periods themselves: construct the Frobenius basisthe canonical set of local solutions of the differential equation at a singular point — one power series per indicial root, with logarithms where roots repeat at each singular point of the Picard–Fuchs operator, solve the connection problem between neighboring bases by transporting solutions numerically along the equation to certified precision, and recognize the resulting constants by integer-relation search — in the ring the geometry's arithmetic dictates, which the K3 and threefold examples above show is decided by the geometry, no longer by the zeta values of the lower rungs.

Open frontier

For polylogarithms there is a complete toolkit — a basis of functions, a symbol calculus, canonical differential equations, mature software at every step — and a graduate student can compute a polylogarithmic amplitude without ever opening the original mathematics literature. For Calabi–Yau integrals no complete analogue of that toolkit exists. There is no exhaustive dictionary of the iterated integrals over threefold periods, no settled canonical basis for a general family, and no algorithm guaranteed to produce a closed form for a given integral. What exists is a growing map: each new integral solved — a threshold coefficient here, a bound-orbit period there — extends it, and occasionally, as with the lemniscatic ring above, adds a region no one had reached. This is the program's current working frontier, and the reason the ladder ends here for now.

Where this shows up in BootLoops

Further reading