WorksMathematics
On the toric Bando–Kasue–Nakajima conjecture
Abstract
The conjecture of Bando, Kasue, and Nakajima asserts that every simply connected Ricci-flat -manifold that is asymptotically locally Euclidean (ALE), that is, asymptotic at infinity to a flat cone , is hyperkähler for one of its orientations. A toric ALE instanton is a Ricci-flat ALE -manifold with an effective isometric action of the -torus; such a manifold is simply connected. We prove the conjecture for the toric ALE instantons whose group at infinity is conjugate in to a subgroup of : each of them is a multi-Eguchi–Hanson metric with collinear centres. In particular, the conjecture holds for toric ALE instantons with . The proof uses a closed formula for toric gravitational instantons, which expresses the metric through Hankel determinants of a positive measure on the fixed points of the torus action. With it, the problem becomes a question about lattice points in a parallelogram determined by the end.
Introduction
A gravitational instanton is a complete, non-compact, oriented, Ricci-flat Riemannian 4-manifold whose curvature decays at least quadratically [26 Def. 1.1]. It is asymptotically locally Euclidean (ALE) if, outside a compact set, it is asymptotic to a flat cone , where is a finite group acting freely on the unit sphere, the group at infinity. The hyperkähler ALE instantons are completely understood. Their group is a finite subgroup of , and for every such Kronheimer constructed and classified them [20, 21]. The simplest examples are the Eguchi–Hanson metric [9], with , and the multi-Eguchi–Hanson metrics of Gibbons and Hawking [14], with .
A conjecture of Bando, Kasue, and Nakajima [6 Rem. (1.8)(3)] asserts that there are no others: every simply connected Ricci-flat ALE 4-manifold should be hyperkähler for one of its two orientations. See also [26 Conj. 11], where it is stated without simple connectivity, as the assertion that the curvature is anti-self-dual for one orientation. Gibbons had conjectured earlier that every ALE instanton is half flat and that the group at infinity of its universal covering lies in [13 Conjecture (1)]. A Ricci-flat metric on a simply connected manifold is half flat exactly when it is hyperkähler for one orientation, so for simply connected manifolds the two formulations agree.
Simple connectivity cannot be dropped: there are free quotients of Gibbons–Hawking metrics that are Ricci-flat Kähler ALE but hyperkähler for neither orientation [34, 32].
The conjecture has been proved by Nakajima [29 Cor. 3.4] and the second author [8 Thm. 6.2] when the manifold is spin and , with the trivial spin structure on the end. Li and Sun proved that a toric ALE instanton that is spin is hyperkähler for one orientation, with no condition on [25 Prop. A.21].
Further cases are known under additional geometric assumptions [23, 28, 12]; without such assumptions the conjecture is open, even for . In this paper we consider toric ALE
2020 Mathematics Subject Classification. Primary 53C25; Secondary 53C26, 83C15.
Key words and phrases. Gravitational instantons, Ricci-flat ALE metrics, toric metrics, Bando–Kasue–Nakajima conjecture, Hankel determinants. instantons, which carry an effective isometric action of the torus (Definition 2.1); for the literature on toric gravitational instantons see [26 §§5, 6, 7.1] and [22, 5].
The group at infinity of a toric ALE instanton is cyclic, of order say, and the cross-section of its end is a lens space. The group is conjugate in to a subgroup of exactly when this lens space is (Lemma 2.6). Every toric ALE instanton is simply connected, whatever its group at infinity (Lemma 2.3), so the conjecture of Bando, Kasue, and Nakajima applies to all of them. Our main result is the following.
Theorem A. A toric ALE instanton (Definition 2.1) whose group at infinity is conjugate in to a subgroup of is a multi-Eguchi–Hanson metric with collinear centres, hyperkähler for one of its orientations.
Theorem A assumes no spin structure, no Hermitian structure, and no topological condition on . See Theorem 5.4 for a more detailed statement. Li and Sun proved the case [25 Prop. A.24], as well as the spin case recalled above.
The proof rests on the solitonic form of Geroch’s conjecture for toric gravitational instantons proved in [3]. Its proof gives the metric in closed form, which allows algebraic and combinatorial methods to be applied: the rod vectors of the finite rods of a non-hyperkähler instanton lie in the interior of a parallelogram in the Lie algebra of the torus that is determined by the end [4]. What is new here is the arithmetic of the lattice points in this parallelogram. When is conjugate in to a subgroup of , the lattice points in its interior lie on one line, too few to accommodate the rod vectors of a non-hyperkähler instanton, and the remaining hyperkähler case is the multi-Eguchi–Hanson metric.
For small groups at infinity the hypothesis of Theorem A holds automatically. For every integer prime to is congruent to or modulo , so every lens space is diffeomorphic to , and Theorem A has the following consequence (Corollary 5.5).
Theorem B. A toric ALE instanton (Definition 2.1) whose group at infinity has order , , , , or is a multi-Eguchi–Hanson metric with collinear centres.
Toric ALE instantons and rod structures
A toric ALE instanton is described by two kinds of data. The continuous data are the positions of the fixed points of the torus action on the boundary of the orbit space. The discrete data are the stabilisers of the action along this boundary, a finite sequence of lattice vectors, which we encode in a chain of integers. The group at infinity and the topology of are determined by the chain. We recall these notions, following [4 §2.1] and [3 §2], and fix the conventions used throughout.
Toric ALE instantons
Let be the 2-torus, its Lie algebra, and the lattice of the elements with , so that . A -basis of is a basis of as a free abelian group. The flat space carries the standard action of the torus by .
Definition 2.1. A toric ALE instanton is a complete, connected, orientable Ricci-flat Riemannian 4-manifold with an effective isometric action of , such that is ALE in the following sense. There are a compact set , a finite group that acts freely on , and a diffeomorphism of onto the complement of a closed ball in such that for some and all , with the norms and covariant derivatives of , where is the Euclidean distance from the origin. The group is the group at infinity.
Remark 2.2 (invariant cross-sections). The end is diffeomorphic to . The chart is not assumed to be compatible with the action, but it can be chosen so on the universal covering of the end, to which the action lifts through a finite covering torus of . By the argument in the proof of [24 Prop. 4.38], which applies the centre-of-mass construction of Grove and Karcher [15] on the end, the lift of can be replaced by a chart that is equivariant for an isomorphism of onto and, after composing it with if necessary, orientation-preserving for a given orientation of . So the end is ALE in the sense of [3 Def. 2.9] for either orientation of . The chart differs from the lift of , on which the deck group acts through , by for some , with the corresponding bounds on derivatives, which the centre-of-mass construction provides. Averaging the function over the deck group gives a -invariant function on that differs from by , and its gradient by . Its large level sets are invariant cross-sections of the end, and beyond each of them the end is a collar .
Lemma 2.3. Let be a toric ALE instanton. Then the action has finitely many fixed points, at least one; is simply connected; and all stabilisers of the action are connected. So , with either orientation, is a simply connected toric gravitational instanton in the sense of [3 Defs. 2.1, 2.2] with an ALE end in the sense of [3 Def. 2.9], and the results of [3, 4] apply to it.
Proof. The fixed points are isolated and lie in a compact set, since the model end contains none, so there are finitely many of them [3 Rem. 2.10], say. The manifold deformation retracts onto a compact -invariant domain that contains them and is bounded by an invariant cross-section of the end, since is a collar of (see Remark 2.2). The Euler characteristic of a closed manifold with a torus action equals that of its fixed-point set, by [19 Ch. II, Thm. 5.5] applied to a Killing field of an invariant metric whose flow is dense in the torus. The double of carries the induced action of , which is smooth for an equivariant collar of . It has fixed points and the Euler characteristic , since is a closed 3-manifold, so . The fundamental group of is finite, since is Ricci-flat with Euclidean volume growth [27, 2], so . Since is connected, orientable, and non-compact, Poincaré duality gives and , which vanishes since and the end of is connected. Hence .
When the stabilisers are connected and the action has a fixed point, simple connectivity follows from [3 Rem. 3.9]. Here connected stabilisers are not assumed. Let be the universal covering, which is finite, with the lifted metric and orientation. The Killing fields of the action lift to complete commuting Killing fields of . For the time- flow of the lift of covers the identity of , so it is a deck transformation. It fixes the points over a fixed point of , where the lifted fields vanish, so it is the identity. Hence the lifted fields generate an effective isometric action of on , whose fixed points are the finitely many points over the fixed points of . Since is simply connected and non-compact, its stabilisers are connected [4 Lemma 2.1]. By Definition 2.1, , so has finite curvature energy , and that of is times as large. So is a gravitational instanton in the sense of [3 Def. 2.2]; it is simply connected and toric, and its action has finitely many fixed points, at least one. Hence the Weyl coordinates of a -basis of map its orbit space homeomorphically onto the closed half-plane [3 Prop. 3.8(ii)].
A deck transformation is an orientation-preserving isometry that commutes with the lifted Killing fields, hence with the action. So it preserves the Gram matrix of the Killing fields and , and it induces an orientation-preserving isometry of the orbit space. Hence is again a harmonic conjugate of , so for a constant . Since has finite order , , so . Since is injective on the orbit space, maps every orbit onto itself, and in particular it fixes the fixed points of the action on . A deck transformation with a fixed point is the identity, so the deck group is trivial, is simply connected, and its stabilisers are connected [4 Lemma 2.1].
Fix orientations of and . For a positively oriented basis of , let be the Gram matrix of the corresponding Killing fields, and let be the dual angles. Since is Ricci-flat and the action has fixed points, the distribution orthogonal to the orbits is integrable, and the function is harmonic on the interior of the orbit space with its quotient metric [3 §2.1]. Where , a harmonic conjugate of gives isothermal coordinates [3 §2.1]. As in [3 §2.1] and [4 §2.1], we choose the harmonic conjugate for which
Then are the Weyl coordinates of the basis, and is determined up to an additive constant. If the basis is replaced by a basis of determinant relative to it, then is unchanged and changes by an additive constant. We use -bases of , except in Section 4. Since is simply connected (Lemma 2.3), the Weyl coordinates of a positively oriented -basis of are global: they map the orbit space homeomorphically onto the closed half-plane , and its interior conformally onto the open half-plane [3 Prop. 3.8(ii)].
The fixed points map to points of the axis , the turning points. They divide the axis into the rods , for , and . Over the open half-plane the orbits are tori, over each rod the stabiliser is a circle subgroup of , and over a turning point it is (Figure 1). The convention (2.1) fixes the direction of the axis, so the numbering of the rods depends on the orientations of and . Over the open half-plane the metric takes the Weyl–Papapetrou form

Figure 1. The orbit space of a toric ALE instanton with turning points, in Weyl coordinates. Over the rod the circle generated by the rod vector collapses, and over a turning point the whole torus collapses to a fixed point. The preimage of the closure of a finite rod is a -sphere .
and Ricci-flatness amounts to a harmonic map equation for and a quadrature for the conformal factor [3 (2.1)–(2.4)].
Rod vectors and chains
2.2. Rod vectors and chains. Let be the area form on with for every positively oriented -basis of . The stabiliser over the rod is the circle subgroup generated by a primitive vector , that is, one that is not of the form with and an integer . It is the rod vector, and it is determined up to sign. At the fixed point over the vectors and form a basis of [31]. So the signs can be chosen, uniquely up to one common sign, such that
Then and are bases with the same determinant, so is an integer multiple of :
The turning points together with the rod vectors form the rod structure of . Since the Weyl coordinates are oriented by (1), the rod vectors and the chain of are those of [3, 4], and the results of these papers apply to as they are stated. Finally, we identify with by the basis , which is positively oriented by (3). Then , , and . The rod vectors are therefore determined by the integers , and we consider such sequences on their own.
Definition 2.4. A chain is a vector , . Its rod vectors are
and we write and , the last rod vector. The intersection form of is the symmetric tridiagonal matrix
and the degrees of are the numbers and of negative and of positive eigenvalues of . We call the degree of .1 The chain of a toric ALE instanton , with orientations of and , is the chain defined by (4), and its degrees are the degrees of .
By induction, the rod vectors of a chain satisfy (3), and . In particular . Our is minus the of [4], and from [4] we use only statements about . The chain determines with its torus action up to equivariant diffeomorphism [31], [30, proof of Thm. 1], and it has a direct topological meaning [4 §2.1]. The preimage of the closure of a finite rod is an embedded 2-sphere $S_j\subset M (Figure 1). Consecutive spheres meet transversally in one fixed point, spheres that are not consecutive are disjoint, the classes of $S_1,\ldots,S_{n-1}$ form a basis of $H_2(M;\mathbb{Z})$, and by {citep}[4 (2.6)]`
So is the intersection form of in the basis , for suitable orientations of the spheres, and the degrees are and . Reversing the orientation of replaces by [4 §2.1] and exchanges the two degrees. Reversing the orientation of replaces by and leaves the degrees unchanged.
Remark 2.5 (orientation conventions). With the opposite convention , the chain assigned to is , so its entries are minus the self-intersection numbers of the spheres, in reverse order. In that convention the chain of the minimal resolution of a cyclic quotient singularity (Section 2.3), with its complex orientation, consists of the entries of the Hirzebruch–Jung continued fraction of , all of which are . This is the usual convention in algebraic geometry [18], [10, §2.6], and the multi-Eguchi–Hanson metrics have the chain in their hyperkähler orientation. We follow the convention of [3, 4], in which they have the chain (Example 2.7).
The end
The two semi-infinite rods determine the end of . We recall the arithmetic needed to describe it. For an integer , two integers and are congruent modulo , (mod ), if divides . A unit modulo is an integer prime to . The units are exactly the integers for which (mod ) for some integer , an inverse of modulo , and the number of units among is Euler’s function . The integers and are always units, and they are the only ones up to congruence exactly when , that is, when . For , for instance, the unit is congruent to neither nor .
For a unit modulo let be the cyclic group generated by , . It acts freely on , and the quotient is the lens space . The lens spaces and are homeomorphic if and only if or (mod ) [7], and then they are diffeomorphic. Since , the coordinates and of the last rod vector are coprime, so that is a unit modulo if .
Lemma 2.6. Let be a toric ALE instanton with the chain , and let be its group at infinity. Then , , the cross-section of the end is diffeomorphic to , and is conjugate in to a subgroup of if and only if (mod ).
Proof. Let be an invariant cross-section of the end (see Remark 2.2). Its orbit space is a compact 1-manifold in the closed half-plane [31 §2]. It is not a circle, since then would be a torus bundle over the circle, whose fundamental group is infinite. So it is an arc with its end points on the axis. Since separates the fixed points from infinity, the arc separates the turning points from infinity in the half-plane, so its end points lie on the rods and , with the stabilisers generated by and , and the stabiliser is trivial in between. If , then , and is diffeomorphic to , whose fundamental group is infinite. Since is finite, . A 3-manifold with such an action is determined up to equivariant diffeomorphism by the lattice and the two stabilisers [31 §2], and comparison with a linear action on a lens space shows that is [30 Lemma 3]. Hence .
No non-trivial element of has the eigenvalue , since acts freely on . Every element of has the eigenvalue , so , and is conjugate in to for a unit . By the classification of lens spaces, or . A generator of rotates two orthogonal planes by the angles and , while an element of rotates two orthogonal planes by opposite angles. So is conjugate in into if and only if , that is, if and only if . □
We say that has the end if the cross-section is diffeomorphic to , without reference to an orientation. For a chain with we likewise call its end. By Lemma 2.6 and the classification of lens spaces, the following three conditions are therefore equivalent: is conjugate in to a subgroup of , the end is with , and (mod ).
Example 2.7. The multi-Eguchi–Hanson metrics with collinear centres are the hyperkähler ALE metrics obtained from the Gibbons–Hawking ansatz with centres on a line in [14], [26 Ex. 4.3], [4 (2.10)]. For the metric is flat, and for it is the Eguchi–Hanson metric [9]. They are toric ALE instantons whose turning points are the centres [3 §11.2]. In their hyperkähler orientation the chain is , and is minus the Cartan matrix of , so that and [4 Rem. 3.6]. With the opposite orientation the chain is , and the two degrees are exchanged. The rod vectors and the end are given in Table 1.
| chain | rod vectors | end | |||
Table 1. The chains used as examples, with the number of turning points, the rod vectors (5), the last rod vector , the end (Lemma 2.6), and the degrees and (Definition 2.4). In the first two rows .
Table 1 collects the data of the chains that serve as examples in this paper: the chains of Example 2.7, the chain of the multi-Eguchi–Hanson metrics with four centres, with the orientation opposite to the hyperkähler one, and the chains and , whose end is not covered by our argument (Example 4.8). The rod vectors follow from (5), the ends from Lemma 2.6 and the classification of lens spaces, and the degrees from the eigenvalues of , or more quickly from Lemma 3.1 below.
Chains and their degrees
The degree of a chain was defined in Definition 2.4 as the number of negative eigenvalues of the intersection form of (6). It can also be read off from the rotation of the rod vectors, and this description identifies with the degree of Li and Sun. By (3), each is obtained from by a clockwise rotation through an angle , followed by a dilation. Here clockwise refers to the orientation of , and the angles are measured in the Euclidean structure of . The total turning of the rod vectors is
the clockwise angle from to swept out by the rod vectors. We also put .
Lemma 3.1. Let be a chain with , with the intersection form of (6), the degrees and of Definition 2.4, and the total turning of (8). Then , so that is non-degenerate and . Moreover , and
is the number of multiples of in the interval . Equivalently, is the number of sign changes in the sequence of first coordinates of the rod vectors, where a vanishing term counts as one sign change between its neighbours. In particular has the sign .
Proof. Let be the leading principal minor of , with . Expanding along the last row gives and , while (5) gives , , and . By induction,
In particular .
Let be the clockwise angle from to , so that . Since , . The angles increase in steps less than . So a vanishing with has neighbours of opposite signs, and two consecutive terms do not vanish. Since and , the number of sign changes in , counted as in the statement, is the number of multiples of in , and . It remains to show that . We use Jacobi’s rule for the signature of a real symmetric matrix, with Gundelfinger’s supplement for a vanishing minor [11 Ch. X, §3, Thm. 2 and Note 1]. For a symmetric tridiagonal matrix with non-zero entries next to the diagonal and non-zero determinant, the leading principal minors have no two consecutive zeros, and a vanishing minor has neighbours of opposite signs, since when . The rules then say that the number of negative eigenvalues is the number of sign changes in this sequence, where a vanishing minor counts as one sign change. By (10) this sequence is , with the same convention for vanishing terms, so . Finally and , so has the sign . □
For the chains of Table 1 the lemma can be checked directly. The rod vectors of the chain have the first coordinates , which change sign twice, so . The chains have the positive first coordinates , so , while the chains have the first coordinates , so .
Remark 3.2 (chains as walks in the Farey graph). The vertices of the Farey graph are the points of , and two fractions and in lowest terms are adjacent if [16 Ch. 1–3]. By (3), consecutive rod vectors form a basis of . So their slopes , with , are fractions in lowest terms, and consecutive slopes are adjacent. Hence a chain is a walk , , in the Farey graph, that is, a sequence of vertices in which consecutive ones are adjacent, from the slope of to that of . The slope decreases along each step on the circle , since is obtained from by a clockwise rotation through an angle in . By Lemma 3.1, is the number of steps from to , , with or , that is, of the steps after the first that cross or reach . For the chain the walk is , and these are the steps from to and from to .
Remark 3.3 (the degree of Li and Sun). For a toric ALE instanton, is the degree that Li and Sun assign to the rod structure of [24 (3.2), Def. 3.11], and is the degree of the rod structure of with the opposite orientation. We use the description of their degree in [4 §2.1], in the orientation convention (1). Each of the lines spanned by the rod vectors is rotated counter-clockwise through an angle in onto the next, and finally onto . Since is obtained from by a clockwise rotation through , the first angles are . The last angle, counter-clockwise from to , is the clockwise angle from to , which is by Lemma 3.1, as by Lemma 2.6. The sum is , and for an ALE end the degree is this sum divided by , minus , that is, . Reversing the orientation of exchanges and (§ 2.2), which gives the second statement.
The closed formula and the diamond inequality
The proof of the main theorem needs one consequence of the analytic theory of toric gravitational instantons: for a toric ALE instanton whose degree satisfies , the interior rod vectors lie in the open parallelogram with the vertices and , the diamond of the end. This inequality is due to [4]. It follows from the closed formula for the metric [3], which expresses the rod vectors through Hankel determinants of a positive measure on the turning points, and from Heine’s formula, which writes these determinants as sums of positive terms. We recall Hankel determinants, state the closed formula, and include the short proof of the inequality for completeness.
Hankel determinants and Heine’s formula
Let be a measure with complex weights on distinct real points . Its moments are , , and its Hankel determinants are
For a set the Heine term is
with empty products equal to 1.
Lemma 4.1 (Heine’s formula). For every ,
Proof. For both sides equal 1. For let be the matrix with the entries , . Then , and the Cauchy–Binet formula gives , where is the submatrix of the rows in . This is a Vandermonde matrix, with . □
Formula (13) goes back to Heine [17]. See also [33 Ch. II] and [3 Lemma 5.1]. In particular, if all weights are positive, then for , and for .
Now let be a positive measure on points , so that . For let
be with the signs of its first weights reversed. By Heine’s formula, is a sum of the positive Heine terms with signs, and this bounds its size. For two points, for example, lies strictly between and .
Lemma 4.2 ([4 Lemma 3.1]). Let be a positive measure on points and . Then , and the following hold.
(i) , and for .
(ii) If , then for .
Proof. By (12), , and . So by (13), is the sum of the positive terms , , with these signs. For every set of atoms has the sign , and for the only set is , with the sign . This proves (i). The triangle inequality gives , with equality only if all signs agree. Let and , and choose a set of atoms. The sets and carry opposite signs, since the first atom is reversed in and the last one is not. Hence the inequality is strict, which proves (ii). □
The closed formula
The closed formula is built from the following functions. Let be points of the axis, a positive measure, and . For and not one of the points put
so that . Let , , and be the measures on the points with the weights , , and , and let
By Heine’s formula, and , since . On the rod , , we have , for , and for , so that there. For put
Theorem 4.3 ([3 Thm. 10.1, (10.1)], in the form of [4 Thm. 2.4]). Let be a toric ALE instanton with turning points and the rod vectors of § 2.2. Then there are an integer with , a positively oriented basis of , a positive measure on the turning points , in the Weyl coordinates of this basis with chosen as in (2.1), and a constant with the following properties, where , , , and are defined by (4.5) and (4.6), and by (4.4) and (4.7), with the Hankel determinants and of the orders and .
(i) With the angles dual to ,
(ii) After a change of the common sign of the rod vectors, which does not change the chain, the components of with respect to are
Theorem 4.3 is the detailed form of the theorem that toric gravitational instantons are gravitational solitons [3 Thm. 1], for an ALE end (for a Pfaffian form of the Belinski–Zakharov solitons see also [1 §§2.3–2.4]). Its hypotheses hold by Lemma 2.3. The soliton data of [3, 4] contain a further number , the weight of an atom at infinity, which vanishes exactly for ALE ends [4 §2.3], so that [4 (2.16)] reduces to (4.6) and (4.8). Items (i) and (ii) are items (ii) and (iii) of [4 Thm. 2.4], and , the order of the Hankel determinants, is the degree index of [3, 4]. Since is positively oriented, its Weyl coordinates differ from those of a positively oriented -basis of by a positive factor and a translation, so the turning points, the rods, and their numbering are those of § 2.1. Item (ii) says that along each rod the generator of the stabiliser is given by Hankel determinants of the measure with some signs reversed. Item (i) of [4 Thm. 2.4] relates the degree index to the degree of Li and Sun, and with Remark 3.3 it identifies the degree index with our degree . The following lemma gives a direct argument from item (ii).
Lemma 4.4. The integer of Theorem 4.3 equals the degree of of Definition 2.4.
Proof. Let
be the normalised Hankel family of [3 §5]. It is the image of under a diagonal linear map with positive entries. By (4.9), and since is positively oriented, is therefore a positive multiple of the determinant of the components, for . In particular by (2.3), so is obtained from by a clockwise rotation through an angle in , followed by a dilation, and these angles add up to [3 Prop. 5.5]. By Lemma 4.2(i), and , so . The argument of the proof of Lemma 3.1, with in place of , shows that the number of sign changes in the sequence , counted as there, is the number of multiples of in , that is, . This sequence has the signs of , so by Lemma 3.1 the number of its sign changes is . Hence . □
The diamond inequality
Recall that is the first coordinate of the last rod vector . Its absolute value is the area of the parallelogram spanned by and , and by Lemma 2.6 it is the order of the group at infinity. Since , the vectors and form a basis of , and we write
for , and for the rod vectors. Recall that . By Lemma 4.4, the order of the Hankel determinants in Theorem 4.3 is . The following comparison of components is taken from the proof of [4 Cor. 3.2(d)].
Lemma 4.5. Let be a toric ALE instanton with the degree , and let be as in Theorem 4.3. Then for ,
Proof. Let be the linear map that sends , with the components with respect to , to . By (4.9), is the vector of the two quotients in (4.12). By Lemma 4.2(i), and . Since is linear, , which is (4.12). □
Definition 4.6. For a chain with the open diamond of its end is the open parallelogram with the vertices and ,
The two descriptions agree by (4.11). The open diamond is convex and invariant under . Since , it is the set where and . The end rod vectors and are vertices of the diamond, where the sum in (4.13) equals . The diamond inequality below concerns only the interior rod vectors .
Proposition 4.7 (the diamond inequality, [4 Cor. 3.2(d)]). Let be a toric ALE instanton with turning points and the degree , and suppose that . Then the interior rod vectors lie in the open diamond of the end:
Proof. Both and lie in . So Lemma 4.2(ii), applied to the orders and , and Lemma 4.5 give and for , that is, . □
In the degrees and excluded in Proposition 4.7 the metric is hyperkähler (Proposition 5.3). For the first quotient in (4.12) is , so , and the second one has absolute value at most by Lemma 4.2, so . Hence , and all rod vectors lie on the segment from to , on the boundary of the diamond. For the chains , for instance, .
Example 4.8. We compare the diamonds of two ends (Figure 2). For we have .

Figure 2. The open diamond (shaded) with the rod vectors (arrows) and its non-zero lattice points (red dots), for two chains of Table 1. The end rod vectors and are vertices of the diamond. (a) For the chain the rod vectors lie on the boundary of the diamond, on the segment from to , and the only non-zero lattice points inside are , joined by a dashed segment, and is not a rod vector. (b) For the chain the diamond contains eight non-zero lattice points on four lines, and all four interior rod vectors lie inside.
The chain of Table 1 has . Here , so the open diamond is the set where . Its only lattice points are and , all on one line. By Lemma 5.2 below, no chain with this end, or with any end , and has all its interior rod vectors in the open diamond.
The chain of Table 1 has the end and the degree , so that . It is invariant under the change that reverses the orientation of (§ 2.2). Here , so the open diamond is the set where . Its non-zero lattice points are , , , and , on four different lines. The interior rod vectors give the sums , so they lie in the open diamond, and the chain satisfies (4.14). So the diamond inequality does not exclude a toric ALE instanton with this chain, and the argument of § 5 says nothing about the end . We show this chain because the chains of toric ALE instantons that are hyperkähler for neither orientation have entries in [25 Lemma A.20], [4 Thm. 2.3], and such a chain with has at least four entries. Shorter chains of the end also satisfy (4.14) with , for instance the chain of Table 1.
Proof of the main theorem
The proof combines the diamond inequality with an elementary fact about lattice points. If , which by Lemma 2.6 means that the group at infinity is conjugate in to a subgroup of , then all lattice points of the open diamond of the end lie on one line. Two consecutive rod vectors form a lattice basis, so they cannot both lie on this line. By Proposition 4.7, the degree of a toric ALE instanton with such an end is therefore or . In these two degrees the metric is a multi-Eguchi–Hanson metric (Proposition 5.3). For this can also be read off from the closed formula (4.8) of Theorem 4.3, which is then the Gibbons–Hawking form of the metric [3 Prop. 11.3].
The lattice lemma
We first determine the lattice points of the open diamond of an end with , and then show that such a diamond cannot contain two consecutive rod vectors.
Lemma 5.1. Let be a chain with , and suppose that with . Then for a primitive vector , and the lattice points of the open diamond are the multiples with . In particular all of them lie on the line .
Proof. Since and , the vector is times , which is primitive. Let be a lattice point of the open diamond, so that . By (21) the numbers and are integers, and . The second coordinate of is , which is an integer, so . Since , this forces . Hence , with . Conversely every such lies in the open diamond. □
For the chains the vector satisfies for all their rod vectors (Table 1), so its slope is adjacent in the Farey graph to the slopes of all rod vectors (Remark 3.2 and Figure 2(a)). In general, the condition says that the slope of is adjacent to the slopes of and , since and . Lemma 5.1 says that in the open diamond only the line is available, which by the opening of this section is too little for two consecutive rod vectors.
Lemma 5.2 (lattice lemma). Let be a chain with and whose interior rod vectors lie in the open diamond . Then . In particular and .
Proof. Suppose that . By Lemma 5.1, and are multiples of the same vector , so , which contradicts (3). The last statement holds since every unit modulo is congruent to (§ 2.3). □
The degrees 0 and
In the two extreme degrees the metric is hyperkähler for one of the orientations, by a theorem of Li and Sun, and the toric hyperkähler ALE instantons are known. The following proposition is [4 Cor. 3.2(c), Rem. 3.6].
Proposition 5.3. Let be a toric ALE instanton with turning points and the degree . Then its chain is , , and is a multi-Eguchi–Hanson metric with collinear centres that is hyperkähler for the orientation opposite to that of . If instead , then the chain is , , and is a multi-Eguchi–Hanson metric with collinear centres that is hyperkähler for the orientation of .
Proof. Let . By Remark 3.3, the degree of Li and Sun of is , and by Lemma 3.1. By the theorem of Li and Sun [24 Thm. 7.4], in the form of [4 Thm. 2.2], is therefore hyperkähler for the orientation of , and is a Gibbons–Hawking metric with collinear centres. The centres are the turning points [3 §11.2], and since the end is ALE, is a multi-Eguchi–Hanson metric [4 Rem. 3.6]. That remark is stated for , and for a Gibbons–Hawking metric with one centre and an ALE end is flat, the multi-Eguchi–Hanson metric with one centre. The chain is by [4 Cor. 3.2(c)], since the degree index of [4] is (Lemma 4.4), and by Table 1. If , then with the opposite orientation has degree , hence reversing the orientation exchanges and (§ 2.2), and the case just treated applies to it. So is a multi-Eguchi–Hanson metric that is hyperkähler for the orientation opposite to that of , the chain of is , and by Table 1. □`
The main theorem
We now combine the diamond inequality, the lattice lemma, and the description of the extreme degrees.
Theorem 5.4. Let be a toric ALE instanton whose group at infinity is conjugate in to a subgroup of . Then the action has exactly fixed points. For one of the orientations of , is a multi-Eguchi–Hanson metric with collinear centres that is hyperkähler for this orientation, and its chain is . For it is the flat , and for it is the Eguchi–Hanson metric.
Proof. By Lemma 2.3, is simply connected, the stabilisers are connected, and the action has fixed points. Fix orientations of and , and let be the chain, , and the degree. By Lemma 2.6, , and the hypothesis says that (§ 2.3). Suppose that . Then , and by Proposition 4.7 the interior rod vectors lie in the open diamond . By Lemma 5.2, , which contradicts the hypothesis. Hence or , and Proposition 5.3 shows that is a multi-Eguchi–Hanson metric with collinear centres, with the chain in its hyperkähler orientation. In both cases . A multi-Eguchi–Hanson metric with one centre is flat, and with two centres it is the Eguchi–Hanson metric. □
The hypothesis on enters only through the lattice lemma.
Corollary 5.5. Let be a toric ALE instanton whose group at infinity has order . Then is a multi-Eguchi–Hanson metric with collinear centres.
Proof. By Lemma 2.6, , and is a unit modulo . For every unit modulo is congruent to (§ 2.3), so , and Theorem 5.4 applies. □
Our argument does not cover . In that case, the end is possible a priori, and for this end the diamond inequality alone does not suffice (Example 4.8). Computer calculations by the authors, which use a reduction to finitely many chains and interval arithmetic and are not part of this paper, indicate that the conclusion of Corollary 5.5 holds for as well.
Computation and the use of an AI assistant
The project was carried out with the help of an AI assistant, Claude (Anthropic), used through the Claude Code environment. The assistant took part in developing the arguments and in writing the text. We also used it to review the literature and to read the drafts critically. Computer calculations were used only for the remark on the order at the end of § 5. The proofs in this paper do not use them. The authors take responsibility for the content.
(Lars Andersson) Department of Mathematics, KTH Royal Institute of Technology, 100 44 Stockholm, Sweden Email address: [email protected]
(Mattias Dahl) Department of Mathematics, KTH Royal Institute of Technology, 100 44 Stockholm, Sweden Email address: [email protected]
(Gustav Nilsson) Mathematisches Institut, Universität Münster, Einsteinstrasse 62, 48149 Münster, Germany Email address: [email protected]
References
- [1]Steffen Aksteiner, Pure spinor factorization of Belinski–Zakharov solitons and regular gravitational instantons, 2026, arXiv:2611.03750v1.arxiv.org/abs/2611.03750
- [2]Michael T. Anderson, On the topology of complete manifolds of nonnegative Ricci curvature, Topology 29 (1990), no. 1, 41–55.DOI
- [3]Lars Andersson, Mattias Dahl, and Gustav Nilsson, On the Geroch conjecture for toric gravitational instantons, 2026, arXiv:2610.41875v1.arxiv.org/abs/2610.41875
- [4]———, Toric gravitational instantons with four fixed points, 2026, arXiv:2611.05333v1.arxiv.org/abs/2611.05333
- [5]Bernardo Araneda and James Lucietti, All toric Hermitian ALE gravitational instantons, J. Geom. Anal. 36 (2026), no. 11, Paper No. 345, arXiv:2510.09291.DOI
- [6]Shigetoshi Bando, Atsushi Kasue, and Hiraku Nakajima, On a construction of coordinates at infinity on manifolds with fast curvature decay and maximal volume growth, Invent. Math. 97 (1989), no. 2, 313–349.DOI
- [7]E. J. Brody, The topological classification of the lens spaces, Ann. of Math. (2) 71 (1960), 163–184.DOI
- [8]Mattias Dahl, The positive mass theorem for ALE manifolds, Mathematics of Gravitation, Part I (Warsaw, 1996) (Piotr T. Chruściel, ed.), vol. 41, Banach Center Publ., no. 1, Polish Acad. Sci. Inst. Math., Warsaw, 1997, pp. 133–142.DOI
- [9]Tohru Eguchi and Andrew J. Hanson, Asymptotically flat self-dual solutions to Euclidean gravity, Phys. Lett. B 74 (1978), no. 3, 249–251.
- [11]F. R. Gantmacher, The theory of matrices. Vol. 1, Chelsea Publishing Co., New York, 1959.
- [12]Partha Ghosh, Mass and rigidity in almost Kähler geometry, 2026, arXiv:2603.08627v2.arxiv.org/abs/2603.08627
- [13]G. W. Gibbons, Gravitational instantons: a survey, Mathematical Problems in Theoretical Physics (Proc. Internat. Conf. Math. Phys., Lausanne, 1979) (K. Osterwalder, ed.), Lecture Notes in Phys., vol. 116, Springer, Berlin, 1980, pp. 282–287.DOI
- [14]G. W. Gibbons and S. W. Hawking, Gravitational multi-instantons, Phys. Lett. B 78 (1978), no. 4, 430–432.
- [15]Karsten Grove and Hermann Karcher, How to conjugate C¹-close group actions, Math. Z. 132 (1973), 11–20.
- [16]Allen Hatcher, Topology of numbers, American Mathematical Society, Providence, RI, 2022.
- [17]Eduard Heine, Handbuch der Kugelfunktionen. Theorie und Anwendungen. Erster Band, second ed., G. Reimer, Berlin, 1878.
- [19]Shoshichi Kobayashi, Transformation groups in differential geometry, Ergebnisse der Mathematik und ihrer Grenzgebiete, vol. 70, Springer-Verlag, New York–Heidelberg, 1972.
- [20]P. B. Kronheimer, The construction of ALE spaces as hyper-Kähler quotients, J. Differential Geom. 29 (1989), no. 3, 665–683.
- [21]⎯⎯⎯⎯, A Torelli-type theorem for gravitational instantons, J. Differential Geom. 29 (1989), no. 3, 685–697.DOI
- [22]Hari K. Kunduri and James Lucietti, On the existence of toric ALE and ALF gravitational instantons, Ann. Henri Poincaré, online first, 2026, arXiv:2604.15159.DOI
- [23]Mingyang Li, On 4-dimensional Ricci-flat ALE manifolds, 2024, arXiv:2304.01609v4.arxiv.org/abs/2304.01609
- [24]Mingyang Li and Song Sun, Gravitational instantons and harmonic maps, 2025, arXiv:2507.15284v1.arxiv.org/abs/2507.15284
- [25]⎯⎯⎯⎯, Compact toric Einstein four-manifolds, 2026, arXiv:2610.10425v2.arxiv.org/abs/2610.10425
- [26]⎯⎯⎯⎯, Geometry of gravitational instantons, 2026, arXiv:2606.10443v1.arxiv.org/abs/2606.10443
- [27]Peter Li, Large time behavior of the heat equation on complete manifolds with nonnegative Ricci curvature, Ann. of Math. (2) 124 (1986), no. 1, 1–21.DOI
- [28]Michael T. Lock and Jeff A. Viaclovsky, Quotient singularities, eta invariants, and self-dual metrics, Geom. Topol. 20 (2016), no. 3, 1773–1806, arXiv:1501.03234.
- [29]Hiraku Nakajima, Self-duality of ALE Ricci-flat 4-manifolds and positive mass theorem, Recent Topics in Differential and Analytic Geometry, Adv. Stud. Pure Math., vol. 18-I, Kinokuniya, Tokyo; Academic Press, Boston, MA, 1990, pp. 385–396.
- [30]Gustav Nilsson, Topology of toric gravitational instantons, Differential Geom. Appl. 96 (2024), Paper No. 102171, arXiv:2301.10212.DOI
- [31]Peter Orlik and Frank Raymond, Actions of the torus on 4-manifolds. I, Trans. Amer. Math. Soc. 152 (1970), no. 2, 531–559.DOI
- [32]Ioana Șuvaina, ALE Ricci-flat Kähler metrics and deformations of quotient surface singularities, Ann. Global Anal. Geom. 41 (2012), no. 1, 109–123, arXiv:1102.2672.arxiv.org/abs/1102.2672
- [33]Gábor Szegő, Orthogonal polynomials, 4th ed., Amer. Math. Soc. Colloquium Publications, vol. 23, American Mathematical Society, Providence, RI, 1975.
- [34]Evan P. Wright, Quotients of gravitational instantons, Ann. Global Anal. Geom. 41 (2012), no. 1, 91–108, arXiv:1102.2442.arxiv.org/abs/1102.2442