ScalingStacks

5.2. The definite triple ω ¯ ϵ [02HL]

Original official author HTML, exact retained edition. Historical TeX conversion verdicts remain unchanged. Cited-edition alignment and mathematical self-containment are not assessed.

Complete original source context · Original author HTML

5.2. The definite triple ω¯ϵ\underline{\omega}_{\epsilon}

We are now going to define an approximately hyperkähler triple 𝝎¯ϵ\bm{\underline{\omega}}_{\epsilon} on the 44–manifold MϵM_{\epsilon}.

For each j=1,…,8j=1,\dots,8 denote by 𝝎¯qj,ϵ\bm{\underline{\omega}}_{q_{j},\epsilon} the hyperkähler triple obtained from the Gibbons–Hawking ansatz (3.3) using the harmonic function

(5.2) hqj=(1+ϵ​λj)+ϵ⁡(mj−2)ρ.h_{q_{j}}=(1+\epsilon\lambda_{j})+\frac{\epsilon(m_{j}-2)}{\rho}.

By abuse of notation we think of 𝝎¯qj,ϵ\bm{\underline{\omega}}_{q_{j},\epsilon} as defined both on the circle bundle H2​mj−4→ℝ3∖B8​ϵ​(0)H^{2m_{j}-4}\rightarrow\mathbb{R}^{3}\setminus B_{8\epsilon}(0) as well as on its quotient by the involution that acts as the simultaneous standard involution on ℝ3\mathbb{R}^{3} and the circle fibres.

Similarly, for each i=1,…,ni=1,\dots,n let 𝝎¯pi,ϵ\bm{\underline{\omega}}_{p_{i},\epsilon} be the hyperkähler triple obtained from the Gibbons–Hawking ansatz using the harmonic function

(5.3) hpi=(1+ϵ​λi)+ϵ​kiρ+ϵ​ℓi.h_{p_{i}}=(1+\epsilon\lambda_{i})+\frac{\epsilon\,k_{i}}{\rho}+\epsilon\,\ell_{i}.

Here λj,λi\lambda_{j},\lambda_{i} and ℓi\ell_{i} are the constants and linear functions appearing in Lemma 4.7. We will assume that ϵ\epsilon is small enough to guarantee that 12<λi,λj<32\tfrac{1}{2}<\lambda_{i},\lambda_{j}<\tfrac{3}{2}.

For all j=1,…,8j=1,\dots,8 let (Mj,𝝎¯Mj)(M_{j},\bm{\underline{\omega}}_{M_{j}}) be a complete DmjD_{m_{j}} ALF space. By Definition 3.6 there exists a compact set K⊂MjK\subset M_{j}, R0>0R_{0}>0 and a diffeomorphism Mj∖K≃H2​mj−4/ℤ2M_{j}\setminus K\simeq H^{2m_{j}-4}/\mathbb{Z}_{2} such that

ϵ2​𝝎¯Mj=𝝎¯qj,ϵ+𝜼¯qj,ϵ\epsilon^{2}\bm{\underline{\omega}}_{M_{j}}=\bm{\underline{\omega}}_{q_{j},\epsilon}+\bm{\underline{\eta}}_{q_{j},\epsilon}

for ρ>ϵ​R0\rho>\epsilon R_{0} with 𝜼¯qj,ϵ=O⁡(ϵ3​ρ−3)\bm{\underline{\eta}}_{q_{j},\epsilon}=O(\epsilon^{3}\rho^{-3}) and similar estimates on the derivatives.

For each i=1,…,ni=1,\dots,n let (Ni,𝝎¯Ni)(N_{i},\bm{\underline{\omega}}_{N_{i}}) be a complete Aki−1A_{k_{i}-1} ALF space. By the classification of ALF spaces of cyclic type [35] the hyperkähler structure on NiN_{i} is explicitly given via the Gibbons–Hawking ansatz starting from a harmonic function on ℝ3\mathbb{R}^{3} with kik_{i} singularities. We can add to this function the smooth harmonic function ϵ2​ℓi\epsilon^{2}\ell_{i}. Over a ball in ℝ3\mathbb{R}^{3} of radius much smaller than ϵ−2\epsilon^{-2} we can regard the resulting hyperkähler structure as a small perturbation of the ALF Aki−1A_{k_{i}-1} hyperkähler structure 𝝎¯Ni\bm{\underline{\omega}}_{N_{i}}. By abuse of notation we denote this perturbed hyperkähler structure with the same symbol 𝝎¯Ni\bm{\underline{\omega}}_{N_{i}}. The advantage of this modification is that now ϵ2​𝝎¯Ni\epsilon^{2}\bm{\underline{\omega}}_{N_{i}} approaches 𝝎¯pi,ϵ\bm{\underline{\omega}}_{p_{i},\epsilon} with a smaller error: by Definition 3.6 there exists a compact set K⊂NiK\subset N_{i}, R0>0R_{0}>0 and a diffeomorphism Ni∖K≃Hki|ℝ3∖BR0N_{i}\setminus K\simeq H^{k_{i}}|_{\mathbb{R}^{3}\setminus B_{R_{0}}} such that

ϵ2​𝝎¯Ni=𝝎¯pi,ϵ+𝜼¯pi,ϵ\epsilon^{2}\bm{\underline{\omega}}_{N_{i}}=\bm{\underline{\omega}}_{p_{i},\epsilon}+\bm{\underline{\eta}}_{p_{i},\epsilon}

with (by scaling) 𝜼¯pi,ϵ=O⁡(ϵ3​ρ−3)\bm{\underline{\eta}}_{p_{i},\epsilon}=O(\epsilon^{3}\rho^{-3}) and similar estimates on the derivatives.

Remark.

The choice of perturbing the gravitational instanton NiN_{i} of type Aki−1A_{k_{i}-1} by adding a linear function on ℝ3\mathbb{R}^{3} can be regarded as an intermediate choice between resolving the singularity of MϵghM^{\textup{gh}}_{\epsilon} by assuming all punctures pip_{i} have weight ki=1k_{i}=1 and the direct gluing of NiN_{i} to MϵghM^{\textup{gh}}_{\epsilon}. While not strictly necessary, the choice of perturbing NiN_{i} by a linear function makes the exposition more uniform. In particular, the closed definite triple we will construct below fails to be hyperkähler by the same amount in a neighbourhood of qjq_{j} and ±pi\pm p_{i}.

We will assume that ϵ\epsilon is chosen so small as to make sure that ϵ​R0≪ρ0\epsilon R_{0}\ll\rho_{0}, where ρ0>0\rho_{0}>0 was fixed in Lemma 4.7. By choosing R0R_{0} larger if necessary we can assume that the harmonic functions hqjh_{q_{j}} and hpih_{p_{i}} in (5.2) and (5.3) are as close to constant functions as we please for ϵ​R0≤ρ≤ρ0\epsilon R_{0}\leq\rho\leq\rho_{0}. Then the hyperkähler triples 𝝎¯qj,ϵ\bm{\underline{\omega}}_{q_{j},\epsilon} and 𝝎¯pi,ϵ\bm{\underline{\omega}}_{p_{i},\epsilon} define metrics gqj,ϵg_{q_{j},\epsilon} and gpi,ϵg_{p_{i},\epsilon} which are uniformly equivalent to g𝕋+ϵ2​θ2g_{\mathbb{T}}+\epsilon^{2}\theta^{2} in the regions ϵ​R0≤ρj,ρi≤2​ρ0\epsilon R_{0}\leq\rho_{j},\rho_{i}\leq 2\rho_{0}. In the rest of the section all norms and covariant derivatives will be computed with respect to this metric.

A crucial observation is that we can take 𝜼¯pi,ϵ\bm{\underline{\eta}}_{p_{i},\epsilon} and 𝜼¯qj,ϵ\bm{\underline{\eta}}_{q_{j},\epsilon} to be exact.

Lemma 5.4.
  1. (i)

    For all i=1,…,ni=1,\dots,n there exists a triple 𝒂¯pi,ϵ\bm{\underline{a}}_{p_{i},\epsilon} of 11–forms on HkiH^{k_{i}} such that

    |∇k𝒂¯pi,ϵ|≤C​ϵ3​(1ρ)2+k|\nabla^{k}\bm{\underline{a}}_{p_{i},\epsilon}|\leq C\epsilon^{3}\left(\frac{1}{\rho}\right)^{2+k}

    and ϵ2​𝝎¯Ni=𝝎¯pi,ϵ+d​𝒂¯pi,ϵ\epsilon^{2}\bm{\underline{\omega}}_{N_{i}}=\bm{\underline{\omega}}_{p_{i},\epsilon}+d\bm{\underline{a}}_{p_{i},\epsilon}.

  2. (ii)

    For all j=1,…,8j=1,\dots,8 there exists a triple 𝒂¯qj,ϵ\bm{\underline{a}}_{q_{j},\epsilon} of ℤ2\mathbb{Z}_{2}–invariant 11–forms on H2​mj−4H^{2m_{j}-4} such that

    |∇k𝒂¯qj,ϵ|≤C​ϵ3​(1ρ)2+k|\nabla^{k}\bm{\underline{a}}_{q_{j},\epsilon}|\leq C\epsilon^{3}\left(\frac{1}{\rho}\right)^{2+k}

    and ϵ2​𝝎¯Mj=𝝎¯qj,ϵ+d​𝒂¯qj,ϵ\epsilon^{2}\bm{\underline{\omega}}_{M_{j}}=\bm{\underline{\omega}}_{q_{j},\epsilon}+d\bm{\underline{a}}_{q_{j},\epsilon}.

Proof.

The proof is identical in the two cases. Set k=kik=k_{i} in case (i) and k=2​mj−4k=2m_{j}-4 in case (ii). In case (ii) we work with ℤ2\mathbb{Z}_{2}–invariant forms on the double cover H2​mj−4H^{2m_{j}-4}.

By scaling we can assume that ϵ=1\epsilon=1. It is enough to prove that every closed 22–form η\eta with η=O⁡(ρ−3)\eta=O(\rho^{-3}) can be written as η=d​a\eta=da with |∇ka|=O⁡(ρ−2−k)|\nabla^{k}a|=O(\rho^{-2-k}).

Since the restriction of HkH^{k} to an exterior domain in ℝ3\mathbb{R}^{3} is diffeomorphic to (R,∞)×Σ(R,\infty)\times\Sigma with Σ\Sigma an homology sphere, we can write η=d​ρ∧α+β\eta=d\rho\wedge\alpha+\beta for some ρ\rho–dependent 11–form α\alpha and 22–form β\beta on Σ\Sigma with |α|+|β|=O⁡(ρ−3)|\alpha|+|\beta|=O(\rho^{-3}).

The condition d​η=0d\eta=0 implies ∂ρβ−dΣ​α=0\partial_{\rho}\beta-d_{\Sigma}\alpha=0. We then define a=−∫ρ∞αa=-\int_{\rho}^{\infty}{\alpha}. The Lemma follows. ∎

By a similar radial integration, Lemma 4.7.(i) implies that in the regions ϵ​R0≤ρj≤2​ρ0\epsilon R_{0}\leq\rho_{j}\leq 2\rho_{0} and ϵ​R0≤ρi≤2​ρ0\epsilon R_{0}\leq\rho_{i}\leq 2\rho_{0}, respectively, we can write

(5.5a) 𝝎¯ϵgh=𝝎¯qj,ϵ+d​𝒂¯qj,ϵgh,𝝎¯ϵgh=𝝎¯pi,ϵ+d​𝒂¯pi,ϵgh\bm{\underline{\omega}}^{\textup{gh}}_{\epsilon}=\bm{\underline{\omega}}_{q_{j},\epsilon}+d\bm{\underline{a}}^{\textup{gh}}_{q_{j},\epsilon},\qquad\bm{\underline{\omega}}^{\textup{gh}}_{\epsilon}=\bm{\underline{\omega}}_{p_{i},\epsilon}+d\bm{\underline{a}}^{\textup{gh}}_{p_{i},\epsilon}
for triples 𝒂¯qj,ϵgh\bm{\underline{a}}^{\textup{gh}}_{q_{j},\epsilon} and 𝒂¯pi,ϵgh\bm{\underline{a}}^{\textup{gh}}_{p_{i},\epsilon} of 11–forms such that
(5.5b) |∇k𝒂¯qj,ϵgh|≤C​ϵ​ρj3−k,|∇k𝒂¯pi,ϵgh|≤C​ϵ​ρi3−k|\nabla^{k}\bm{\underline{a}}^{\textup{gh}}_{q_{j},\epsilon}|\leq C\epsilon\rho_{j}^{3-k},\qquad|\nabla^{k}\bm{\underline{a}}^{\textup{gh}}_{p_{i},\epsilon}|\leq C\epsilon\rho_{i}^{3-k}

for k=0,1,2,3k=0,1,2,3.

Now, let χqj\chi_{q_{j}} and χpi\chi_{p_{i}} be cut-off functions with the following properties:

(5.6) χqj≡1 for ρj≤ϵ25,χqj≡0 for ρj≥2ϵ25,|∇χqj|≤Cρj−1,χpi≡1 for ρi≤ϵ25,χpi≡0 for ρi≥2ϵ25,|∇χpi|≤Cρi−1.\begin{gathered}\chi_{q_{j}}\equiv 1\text{ for }\rho_{j}\leq\epsilon^{\frac{2}{5}},\qquad\chi_{q_{j}}\equiv 0\text{ for }\rho_{j}\geq 2\epsilon^{\frac{2}{5}},\qquad|\nabla\chi_{q_{j}}|\leq C\rho_{j}^{-1},\\ \chi_{p_{i}}\equiv 1\text{ for }\rho_{i}\leq\epsilon^{\frac{2}{5}},\qquad\chi_{p_{i}}\equiv 0\text{ for }\rho_{i}\geq 2\epsilon^{\frac{2}{5}},\qquad|\nabla\chi_{p_{i}}|\leq C\rho_{i}^{-1}.\\ \end{gathered}

We finally define a triple of closed 22–forms on MϵM_{\epsilon} by

(5.7) 𝝎¯ϵ={ϵ2​𝝎¯Mjif ​ρj≤ϵ25,𝝎¯qj,ϵ+d⁡(χqj​𝒂¯qj,ϵ+(1−χqj)​𝒂¯qj,ϵgh)if ​ϵ25≤ρj≤2​ϵ25,ϵ2​𝝎¯Niif ​ρi≤ϵ25,𝝎¯pi,ϵ+d⁡(χpi​𝒂¯pi,ϵ+(1−χpi)​𝒂¯pi,ϵgh)if ​ϵ25≤ρi≤2​ϵ25,𝝎¯ϵghif ​ρj≥2​ϵ25​ and ​ρi≥2​ϵ25​ for all ​i,j.\bm{\underline{\omega}}_{\epsilon}=\begin{cases}\epsilon^{2}\bm{\underline{\omega}}_{M_{j}}&\mbox{if }\rho_{j}\leq\epsilon^{\frac{2}{5}},\\ \bm{\underline{\omega}}_{q_{j},\epsilon}+d\left(\chi_{q_{j}}\,\bm{\underline{a}}_{q_{j},\epsilon}+(1-\chi_{q_{j}})\,\bm{\underline{a}}^{\textup{gh}}_{q_{j},\epsilon}\right)&\mbox{if }\epsilon^{\frac{2}{5}}\leq\rho_{j}\leq 2\epsilon^{\frac{2}{5}},\\ \epsilon^{2}\bm{\underline{\omega}}_{N_{i}}&\mbox{if }\rho_{i}\leq\epsilon^{\frac{2}{5}},\\ \bm{\underline{\omega}}_{p_{i},\epsilon}+d\left(\chi_{p_{i}}\,\bm{\underline{a}}_{p_{i},\epsilon}+(1-\chi_{p_{i}})\,\bm{\underline{a}}^{\textup{gh}}_{p_{i},\epsilon}\right)&\mbox{if }\epsilon^{\frac{2}{5}}\leq\rho_{i}\leq 2\epsilon^{\frac{2}{5}},\\ \bm{\underline{\omega}}^{\textup{gh}}_{\epsilon}&\mbox{if }\rho_{j}\geq 2\epsilon^{\frac{2}{5}}\mbox{ and }\rho_{i}\geq 2\epsilon^{\frac{2}{5}}\mbox{ for all }i,j.\end{cases}
Remark.

When ki=1k_{i}=1 there is no need to glue in an A0A_{0} ALF space (ℝ4\mathbb{R}^{4} endowed with the Taub–NUT metric), since 𝝎¯ϵgh\bm{\underline{\omega}}^{\textup{gh}}_{\epsilon} already extends smoothly over ±pi\pm p_{i}. It is however useful (we will use this in setting up the analysis for the deformation problem) to think of a rescaled Taub–NUT space localised around ±pi\pm p_{i}.

5.2.1. The error

We conclude this section by quantifying the failure of 𝝎¯ϵ=(ωϵ1,ωϵ2,ωϵ3)\bm{\underline{\omega}}_{\epsilon}=(\omega^{1}_{\epsilon},\omega^{2}_{\epsilon},\omega^{3}_{\epsilon}) to define a hyperkähler structure. Since by construction d​ωϵi=0d\omega^{i}_{\epsilon}=0 for all i=1,2,3i=1,2,3, we only have to check that 𝝎¯ϵ\bm{\underline{\omega}}_{\epsilon} is a definite triple and estimate the difference between the associated intersection matrix and the identity.

In the regions ρj≤ϵ25\rho_{j}\leq\epsilon^{\frac{2}{5}}, ρi≤ϵ25\rho_{i}\leq\epsilon^{\frac{2}{5}} and when ρi,ρj≥2​ϵ25\rho_{i},\rho_{j}\geq 2\epsilon^{\frac{2}{5}} for all i,ji,j the triple 𝝎¯ϵ\bm{\underline{\omega}}_{\epsilon} defines a genuine hyperkähler structure. In the transition regions ϵ25≤ρj≤2​ϵ25\epsilon^{\frac{2}{5}}\leq\rho_{j}\leq 2\epsilon^{\frac{2}{5}} and ϵ25≤ρi≤2​ϵ25\epsilon^{\frac{2}{5}}\leq\rho_{i}\leq 2\epsilon^{\frac{2}{5}} we have, respectively,

𝝎¯ϵ−𝝎¯qj,ϵ=O⁡(ϵ2−15),𝝎¯ϵ−𝝎¯pi,ϵ=O⁡(ϵ2−15).\bm{\underline{\omega}}_{\epsilon}-\bm{\underline{\omega}}_{q_{j},\epsilon}=O(\epsilon^{2-\frac{1}{5}}),\qquad\bm{\underline{\omega}}_{\epsilon}-\bm{\underline{\omega}}_{p_{i},\epsilon}=O(\epsilon^{2-\frac{1}{5}}).

Since 𝝎¯qj,ϵ\bm{\underline{\omega}}_{q_{j},\epsilon} and 𝝎¯pi,ϵ\bm{\underline{\omega}}_{p_{i},\epsilon} are hyperkähler triples, we conclude that 𝝎¯ϵ\bm{\underline{\omega}}_{\epsilon} is a definite triple for ϵ\epsilon sufficiently small.

Let μϵ\mu_{\epsilon}, gϵg_{\epsilon} and QϵQ_{\epsilon} be the volume form, metric and intersection matrix associated to the definite triple 𝝎¯ϵ\bm{\underline{\omega}}_{\epsilon} as in Section 2. Using Lemma 5.4, (5.5), (5.6) and the definition of 𝝎¯ϵ\bm{\underline{\omega}}_{\epsilon} we calculate that

(5.8) |Qϵ−id|≤C​ϵ2−15|Q_{\epsilon}-\text{id}|\leq C\epsilon^{2-\frac{1}{5}}

in every transition region ϵ25≤ρj≤2​ϵ25\epsilon^{\frac{2}{5}}\leq\rho_{j}\leq 2\epsilon^{\frac{2}{5}}, j=1,…,8j=1,\dots,8, and ϵ25≤ρi≤2​ϵ25\epsilon^{\frac{2}{5}}\leq\rho_{i}\leq 2\epsilon^{\frac{2}{5}}, i=1,…,ni=1,\dots,n. Outside the transition regions Qϵ≡idQ_{\epsilon}\equiv\text{id}.

Original mathematics by the credited authors. Source-backed reader collection; mathematical self-containment is not assessed.