
conjectured that such an Einstein metric exists on HPm+1♯HPm+1 for all m≥2, as indicated
by numerical evidence provided in [PP86], [B¨oh98] and [DHW13].
Some well-known Einstein metrics are realized as integral curves to the cohomogeneity one
Einstein equation. For example, the standard sphere metric, the sine cone over Jensen’s sphere,
and the quaternionic-K¨ahler metric on HPm+1 are represented by integral curves to the cohomo-
geneity one system. Furthermore, the cone solution is an attractor to the system. It was realized
in [B¨oh98] that the winding of integral curves around the cone solution plays an important role
in the existence problem described above. To investigate the winding, one studies a quantity
(denoted as ♯Cw(¯
h) in [B¨oh98]) that is assigned to each local solution that does not globally
define a complete Einstein metric on HPm+1♯HPm+1. From the point of view of geometry, the
quantity records the number of times that the principal orbit becomes isoparametric while its
mean curvature remains positive. In general, an estimate for ♯Cw(¯
h) can be obtained from the
linearization along the cone solution. For m= 1, the estimate is good enough to prove the global
existence. This is not the case, however, if m≥2. For higher dimensional cases, it is from the
global analysis of the system that we obtain a further estimate for ♯Cw(¯
h) and we prove the
following existence theorem.
Theorem 1.1. On each HPm+1♯HPm+1 with m≥2, there exists at least one positive Einstein
metric with G/K=S4m+3 as its principal orbit.
Numerical studies in [B¨oh98] and [DHW13] indicate that there exists another Einstein metric
on HPm+1♯HPm+1 with m≥2. Based on Theorem 1.1, an estimate for ♯Cw(¯
h) in a limiting
subsystem (essentially obtained from the linearization along the cone solution) helps us propose
a criterion to check the existence of the second Einstein metric. Let nbe the dimension of G/K.
Such a criterion only depends on n(or m).
Theorem 1.2. Let θΨbe the solution to the following initial value problem:
dθ
dη =n−1
2ntanh η
nsin(2θ) + 2
ns(2m+ 1)(2m+ 2)(2m+ 3)
(2m+ 3)2+ 2m, θ(0) = 0.(1.2)
Let Ω = lim
η→∞ θΨ. For m≥2, there exist at least two positive Einstein metrics on HPm+1♯HPm+1
if Ω<3π
4.
The upper bound for Ω in Theorem 1.2 is not sharp. Although it is difficult to solve the initial
value problem (1.2) explicitly, one can use the Runge–Kutta 4-th order algorithm to approximate
Ω. Since the RHS of (1.2) does not vanish at η= 0, the initial Runge–Kutta step is well-defined.
Our numerical study shows that Ω <3π
4for integers m∈[2,100].
We also look into the case where G/Kcompletely collapses at two ends of a compact manifold.
In that case, the cohomogeneity one space is S4m+4. No new Einstein metric is found on S4m+4
for m≥2. For m= 1, however, we obtained a non-standard positive Einstein metric on S8.
Such a metric is inhomogeneous by the classification in [Zil82].
Theorem 1.3. There exists a non-standard Sp(2)Sp(1)-invariant positive Einstein metric ˆgS8
on S8.
It is worth mentioning that all new solutions found are symmetric. Metrics that are repre-
sented by these solutions all have a totally geodesic principal orbit.
This article is structured as follows. In Section 2, we present the dynamical system for
positive Einstein metrics of cohomogeneity one with G/Kas the principal orbit. Then we apply
a coordinate change that makes the cohomogeneity one Ricci-flat system serve as a limiting
subsystem. Initial conditions and terminal conditions are transformed into critical points of
the new system. The new system admits Z2-symmetry. By a sign change, one can transform
initial conditions into terminal conditions. Hence the problem of finding globally defined positive
Einstein metrics boils down to finding heteroclines that join two different critical points.
2