Maximal surfaces: existence, uniqueness, deformations

Chapter IV: Maximal surfaces and AdS 3-manifolds

4.3 Maximal surfaces: existence, uniqueness, deformations

With Proposition 4.2.1 in hand, we prove the main theorems. For convenience we assume the twist parameter is zero—the proof has no dependence on it. The existence result is immediate from the next proposition.

Proposition 4.3.1. The functionalE =E𝜌

1, 𝜌

2 : T (Γ) → Ris proper if and only if 𝜌1almost strictly dominates𝜌

2.

Indeed, by Proposition 4.2.1, properness implies the existence of a maximal space- like immersion. Thus, the proof of the existence result reduces to showing that almost strict domination implies properness.

Mapping class groups

Preparing for the proof of Theorem 4A, we review some Teichmüller theory. Set Diff⁺(Σ) to be the group of 𝐶^∞ orientation preserving diffeomorphisms of Σ, equipped with the 𝐶^∞ topology. Denote by Diff⁺₀(Σ) the normal subgroup con- sisting of maps that are isotopic to the identity. The mapping class group of Σ is defined

MCG(Σ) :=𝜋

0(Diff⁺(𝑆)) =Diff⁺(𝑆)/Diff⁺₀(𝑆). Given a collection of boundary lengthsℓ = (ℓ

1, . . . , ℓ_𝑛) ∈ R^𝑛, we can consider the Teichmüller space Tℓ(Γ) of surfaces with punctures and geodesic boundary with lengths determined by (ℓ

1, . . . , ℓ_𝑛). This is not a relative representation space but a union of them. The mapping class group acts on Teichmüller space by pulling back classes of hyperbolic metrics:

[𝜑] · [𝜇] ↦→ [𝜑^∗𝜇]. (4.13) The action of the mapping class group is properly discontinuous and the quotient is the moduli space of hyperbolic surfaces.

There is a wealth of metrics on Teichmüller space, and the mapping class group acts by isometries on a number of them. In the work below, we set 𝑑_T to be any metric distance function on Teichmüller space on which the mapping class group acts by isometries. One example is the Teichmüller distance.

Selecting a basepoint𝑧 ∈Σ, there is an action of the mapping class group on𝜋

1(Σ, 𝑧), which identifies with Γ. Given 𝜑 ∈Diff⁺(Σ) such that𝜑(𝑧) = 𝑧, the induced map 𝜑_∗ :𝜋

1(Σ, 𝑧) → 𝜋

1(Σ, 𝑧)is an automorphism. While a general𝜑may not fix𝑧, one can find a different 𝜑

1 ∈ Diff⁺(Σ) that does fix𝑧and is isotopic to 𝜑. The isotopy gives rise to a path𝛾

1from𝜑(𝑧)to𝑧. If𝜑

2∈Diff⁺(Σ)also fixes𝑧and is isotopic to 𝜑, then we get another path𝛾

2from𝜑(𝑧)to𝑧. The automorphisms(𝜑

1)_∗ and(𝜑

2)_∗ of 𝜋

1(Σ, 𝑧) differ by the inner automorphism corresponding to the conjugation by the class of𝛾

1·𝛾

2. This association furnishes an injective homomorphism from MCG(Σ) →Out(𝜋

1(Σ)) =Aut(𝜋

1(Σ))/Inn(𝜋

1(Σ)).

Through this injective mapping, the mapping class group acts on representation spaces: we can precompose a representative of a representation with a representative of an element in Out(𝜋

1(Σ)), and then take the corresponding equivalence class.

On a Teichmüller space, this agrees with the action (4.13). For this action, we use the similar notation

[𝜑] · [𝜌] ↦→ [𝜑^∗𝜌] =[𝜌◦𝜑_∗].

Note that it does not in general preserve relative representation spaces: a mapping class may permute the punctures. Finally, we record that MCG(Σ)acts equivariantly with respect to

• length spectrum: for a representation 𝜌 and [𝜑] ∈ MCG(Σ), ℓ(𝜑^∗𝜌(𝛾)) = ℓ(𝜌(𝜑_∗(𝛾))). If 𝜇 is a hyperbolic metric we set ℓ_𝜇(𝛾) to be the 𝜇-length of the geodesic representative of 𝛾 in (Σ, 𝜇). A restatement of the above is ℓ_𝜑∗𝜇(𝛾) =ℓ_𝜇(𝜑(𝛾)).

• Energy of harmonic maps: if 𝑓 : (Σ˜,𝜇˜) → (𝑋 , 𝜈) is a 𝜌-equivariant map, then

𝑒(𝜇, 𝑓_𝜇) =𝑒(𝜑^∗𝜇, 𝑓_𝜇◦𝜑). (4.14) Furthermore, if 𝑓 is harmonic, then 𝑓 ◦𝜑 : (Σ˜,𝜑˜^∗𝜇˜) → (𝑋 , 𝜈) is a 𝜑^∗𝜌- equivariant harmonic map.

Existence of maximal surfaces

Suppose a simple closed curve 𝜉 is either a geodesic boundary component or a horocycle in a hyperbolic surface. If 𝜉 is a boundary, let𝐶_𝑑(𝜉) denote the collar around𝜉 consisting of points with distance to𝜉 at most𝑑. If it is a horocycle, put 𝐶_𝑑(𝜉) to be the union of the 𝑑-collar and the enclosed cusp (both are defined for suitably small𝑑).

We now begin the proof of properness. Suppose 𝜌

1almost strictly dominates 𝜌

2. We assume there is a single peripheral curve—there are no substantial changes in the case of many cusps—and we set𝜉to be either the geodesic boundary component of the convex core ofH/𝜌

1(Γ), or a deep horocycle ofH/𝜌

1(Γ), depending on whether the monodromy is hyperbolic or parabolic. Let𝑔be an optimal(𝜌

1, 𝜌

2)-equivariant map. If the image of the peripheral curve under 𝜌

1and𝜌

2is hyperbolic, then𝑔has constant speed on the boundary components. Note that𝑔◦ℎ_𝜇is𝜌

2-equivariant, and in the hyperbolic case, converges at∞ to a projection onto the geodesic at infinity in the sense of Definition 4.2.13. Thus, from Lemma 4.2.14 we have

E (𝜇) =

∫

Σ

𝑒(𝜇, ℎ_𝜇) −𝑒(𝜇, 𝑓_𝜇)𝑑𝐴_𝜇 ≥

∫

Σ

𝑒(𝜇, ℎ_𝜇) −𝑒(𝜇, 𝑔◦ℎ_𝜇)𝑑𝐴_𝜇. One of the main advantages of replacing 𝑓_𝜇with𝑔◦ℎ_𝜇 is that the integrand is now non-negative. Moving toward the proof of properness, let ( [𝜇_𝑛])^∞

𝑛=1 ⊂ T (Γ) be such that

E (𝜇_𝑛) ≤𝐾

for all 𝑛, for some large 𝐾. It suffices to show [𝜇_𝑛] converges in T (Γ) along a subsequence. Structurally, our proof is similar to that of the classical existence result for minimal surfaces in hyperbolic manifolds (see [SY79] and also the more modern paper [GW07]): we first prove 1) compactness in moduli space, and then 2) extend to compactness in Teichmüller space.

Toward 1), we show there is a𝛿 > 0 such that for all non-peripheral simple closed curves inΓ, we haveℓ_𝜇

𝑛(𝛾) ≥𝛿. We argue by contradiction: suppose this is not the case, so that there is a sequence of non-peripheral simple closed curves(𝛾_𝑛)^∞

𝑛=1 ⊂ Γ such that

ℓ_𝑛 :=ℓ_𝜇

𝑛(𝛾_𝑛) →0

as 𝑛 → ∞. By the regular collar lemma, we know that in (Σ, 𝜇_𝑛), the geodesic representing𝛾_𝑛is enclosed by a collar𝐶_𝑛 of width𝑤_𝑛 = arcsinh( (sinhℓ_𝑛)⁻¹). Up to a conjugation of the holonomy,𝐶_𝑛is conformally equivalent to{𝑥+𝑖 𝑦 ∈C: 0≤ 𝑥 ≤ 𝑤_𝑛,0≤ 𝑦 ≤ 1}with the lines{𝑦=0}and{𝑦=1} identified.

Let 𝑓_𝑛= 𝑓_𝜇

𝑛,ℎ_𝑛= ℎ_𝜇

𝑛. Henceforward, conformally modify the metric 𝜇_𝑛to be flat in𝐶_𝑛. Our control over the energy functional depends on our knowledge of the local Lipschitz constant of the map 𝑔. This in turns relies on the image of the harmonic mapℎ_𝑛, in particular how far it takes𝐶_𝑛into𝐶_𝑑(𝜉). For all𝑥 ,0< 𝑡 < 𝑑in question, set 𝐴^𝑡

𝑥 ={𝜃 ∈ {𝑥} ×𝑆¹: ℎ_𝑛(𝑥 , 𝑦) ∉𝐶_𝑡(𝜉)}. We also write Lip(𝑔, 𝑡) = max

𝑥∈𝐶(H/𝜌

1(Γ))\𝐶_𝑡(𝜉)Lip𝑥(𝑔). We have the inequalities

E (𝜇_𝑛) (𝜎_𝑛) ≥

∫

𝐶_𝑛

𝑒(𝜇_𝑛, ℎ_𝑛) −𝑒(𝜇_𝑛, 𝑔◦ℎ_𝑛)𝑑𝐴_𝜇

𝑛

=

∫ ^𝑤𝑛

0

∫ ₁

0

𝑒(𝜇_𝑛, ℎ_𝑛) −𝑒(𝜇_𝑛, 𝑔◦ℎ_𝑛)𝑑𝑦 𝑑𝑠

≥ 𝑤_𝑛min

𝑥

∫

𝐴^𝑡_𝑥

𝑒(𝜇_𝑛, ℎ_𝑛) −𝑒(𝜇_𝑛, 𝑔◦ℎ_𝑛)𝑑𝑦

≥ 𝑤_𝑛min

𝑥

(1−Lip(𝑔, 𝑡)²)

∫

𝐴^𝑡_𝑥

𝑒(𝜇_𝑛, ℎ_𝑛)𝑑𝑦

≥ 𝑤_𝑛 2 min

𝑥

(1−Lip(𝑔, 𝑡)²)

∫

𝐴^𝑡_𝑥

𝜕 ℎ_𝑛(𝑥 , 𝑦)

𝜕 𝑦

2

𝑑𝑦

≥ 𝑤_𝑛 2 min

𝑥

(1−Lip(𝑔, 𝑡)²) (ℓ(A^𝑡𝑥))⁻¹ ∫

𝐴^𝑡_𝑥

𝜕 ℎ_𝑛(𝑥 , 𝑦)

𝜕 𝑦 𝑑𝑦

₂

≥ 𝑤_𝑛 2 min

𝑥

(1−Lip(𝑔, 𝑡)²)∫

𝐴^𝑡_𝑥

𝜕 ℎ_𝑛(𝑥 , 𝑦)

𝜕 𝑦 𝑑𝑦

₂ .

Here ℓ(𝐴^𝑡

𝑥) is the length of the (possibly broken) segment 𝐴_{𝑥 ,𝑡}, which is clearly bounded above by 1. Fuchsian representations have discrete length spectrum, and hence there is a𝜅 > 0 such thatℓ(𝜌

1(𝛾)) ≥ 𝜅for all𝛾 ∈ Γ. Lemma 4.3.2. Set𝑡 =𝑑/2. Then for any𝑥, the inequality

∫

𝐴^𝑑/_𝑥²

|𝜕 ℎ_𝑛(𝑥 , 𝑦)

𝜕 𝑦

𝑑𝑦 ≥ min{𝑑/2, 𝜅} =: 𝑘 holds.

Proof. Let 𝛼 be any core curve of the form {𝑥} × 𝑆¹. If ℎ_𝑛(𝛼) always remains outside𝐶_𝑑/

2(𝜉), then

∫

𝐴^𝑡_𝑥

|𝜕 ℎ_𝑛(𝑥 , 𝑦)

𝜕 𝑦 𝑑𝑦=

∫ ₁

0

𝜕 ℎ_𝑛(𝑥 , 𝑦)

𝜕 𝑦

𝑑𝑦≥ ℓ(𝜌

1(𝛾_𝑛)) ≥𝜅 . Thus, we assumeℎ_𝑛(𝛼) intersects𝐶_𝑑/

2(𝜉). We parametrize ℎ_𝑛(𝑥 ,·)by arc length, so that the integral in question measures the hyperbolic length on H/𝜌

1(Γ) of the segment of ℎ_𝑛(𝛼) that does not enter𝐶_𝑑/

2(𝜉). If this length is ever less than𝑑/2, then the curveℎ_𝑛(𝛼)is contained in𝐶_𝑑(𝜉). Phrased differently, it is a simple closed curve contained in an embedded cylinder. There is only one such homotopy class of curves, namely the homotopy class of the boundary geodesic. This situation is impossible, sinceℎ_𝑛(𝛼)is homotopic to a non-peripheral simple closed curve, and

hence the length is at least𝑑/2. □

Returning to the inequalities above, we now have E (𝜇_𝑛) ≥ 𝑤_𝑛

2

(1−Lip(𝑔, 𝑑/2)²)𝑘².

We thus findE (𝜇_𝑛) → ∞as𝑛→ ∞, which violates the uniform upper bound.

The lower bound on the length spectrum shows the metrics (𝜇_𝑛)^∞

𝑛=1 satisfy Mum- ford’s compactness criteria [Mum71] and hence project under the action of the mapping class group to a compact subset of the moduli space.

Now we promote to compactness in Teichmüller space. By compactness in moduli space, after passing to a subsequence of the𝜇_𝑛’s, there exists a sequence of mapping class group representatives (𝜓_𝑛)^∞

𝑛=1 such that 𝜓^∗

𝑛𝜇_𝑛 converges as 𝑛 → ∞ to a hyperbolic metric𝜇_∞.

Lemma 4.3.3. Upon passing to a further subsequence, ( [𝜓_𝑛])^∞

𝑛=1 ⊂ MCG(Σ) is eventually constant.

The main thrust of the proof is to show that for each non-peripheral 𝛾 ∈ Γ,ℓ(𝜌

1◦ 𝜓_𝑛(𝛾))is uniformly bounded above in𝑛. Our proof of this claim is by contradiction and similar in nature to the argument above. Suppose there is a class of curves𝛾 ∈Γ such thatℓ(𝜌

1◦𝜓_𝑛(𝛾)) → ∞as𝑛 → ∞. Modify the notation: setℎ_𝑛 =ℎ_𝜇

𝑛◦𝜓_𝑛and 𝑓_𝜇

𝑛 ◦𝜓_𝑛. These are harmonic for𝜓^∗

𝑛𝜌

1and𝜓^∗

𝑛𝜌

2respectively, and have the correct boundary behaviour according to [Sag19, Theorem 1.1]. From (4.14), we also have the equality

E𝜌

1, 𝜌

2(𝜇_𝑛) =E𝜓_𝑛^∗𝜌

1,𝜓^∗_𝑛𝜌

2(𝜓^∗

𝑛𝜇_𝑛). On each (Σ, 𝜓^∗

𝑛𝜇_𝑛), there is a collar𝐶_𝑛 of finite width𝑤_𝑛 around 𝛾, and the𝐶_𝑛’s converge to some collar𝐶_∞in (Σ, 𝜇_∞)of width𝑤_∞ < ∞. As before, we perturb to a flat metric and parametrize𝐶_𝑛by [0, 𝑤_𝑛] ×𝑆¹. We note that 𝑔is (𝜓^∗

𝑛𝜌

1, 𝜓^∗

𝑛𝜌

2)- equivariant. Therefore, we can apply our previous reasoning to see that for𝑛large enough, any small enough𝑥, and𝑡 ∈ [0, 𝑤_𝑛],

E𝜌

1, 𝜌

2(𝜇_𝑛) =E𝜓^∗_𝑛𝜌

1,𝜓^∗_𝑛𝜌

2(𝜓^∗

𝑛𝜇_𝑛) ≥ 𝑤_∞ 4 min

𝑥

(1−Lip(𝑔, 𝑡)²)∫

𝐴^𝑡_𝑥

𝜕 ℎ_𝑛(𝑥 , 𝑦)

𝜕 𝑦 𝑑𝑦

₂ , (4.15) where 𝐴^𝑡

𝑥is defined as above. This leads us to the analogue of Lemma 4.3.2.

Lemma 4.3.4. Set𝑡 =𝑑/2. Then, independent of the choice of𝑥,

∫

𝐴^𝑑/_𝑥²

𝜕 ℎ_𝑛(𝑥 , 𝑦)

𝜕 𝑦

𝑑𝑦 → ∞ as𝑛→ ∞.

Proof. Choose any simple closed curve𝛼of the form{𝑥} ×𝑆¹. Parametrizeℎ_𝑛(𝑥 ,·) by arc length so that the integral above returns the length of the segment ofℎ_𝑛(𝛼) that does not enter𝐶_𝑑/

2(𝜉). If ℎ_𝑛(𝛼) does not enter𝐶_𝑑/

2(𝜉), then we can argue as we did in the proof of Lemma 4.3.2, bounding the length below by theℓ(𝜓^∗

𝑛𝜌

1(𝛾)), which blows up. Thus, we restrict our discussion to 𝛼 such that ℎ_𝑛(𝛼) intersects 𝐶_𝑑/

2(𝜉). Let us first assume there is a positive integer 𝐾 such that every ℎ_𝑛(𝛼) enters𝐶_𝑑/

2(𝜉) at most𝐾 times. Of course, ℎ_𝑛(𝛼) cannot live entirely in 𝐶_𝑑/

2(𝜉), for then it lies in the wrong homotopy class. We construct a new curve as follows:

if we choose a basepoint not within𝐶_𝑑/

2(𝜉), then every timeℎ_𝑛(𝛼)enters𝐶_𝑑/

2(𝜉), there is a corresponding point at which it exits. We plan to erase the segment of ℎ_𝑛(𝛼) that connects these two points and replace it with the one of the two possible paths on𝜕𝐶_𝑑/

2(𝜉), say,𝛽

1and𝛽

2. Ifℎ_𝑛(𝛼)spends some time going in a path along the boundary circle when it is entering or when it is exiting (we count just touching

it once as an exit), we connect the endpoint of the entrance path to the starting point of the exiting path and concatenate with the path going along the boundary.

The question is: which path to take? We argue there is a choice so that the homotopy class of the new curve is the same as that ofℎ_𝑛(𝛼). To this end, orient the boundary circle so that there is a “left” and a “right” path. Arbitrarily choose one of the two paths connecting the entrance and the exit point, say, 𝛽

1, and consider the loop in the cylinder obtained by concatenating this path with the piece ofℎ_𝑛(𝛼) ∩𝐶_𝑑/

2(𝜉) that connects the endpoints. This is a simple closed curve in a cylinder, and hence there are three possible homotopy classes: trivial, non-trivial and left oriented, and non-trivial and right oriented. If the class is trivial, then using this path 𝛽

1 will work. If it is non-trivial and left oriented, then 𝛽

1must be the right oriented path.

Thus, if we use the opposite path𝛽

2, it will cancel the homotopy class, and therefore the new path will indeed be homotopic toℎ_𝑛(𝛼). The right oriented case is similar.

Each replacement adds at most ℓ(𝜕𝐶_𝑑/

2(𝜉)) to the length of the path. Thus, the length of the new path is (quite crudely) bounded above by

ℓ_𝑑/

2(ℎ_𝑛(𝛼)) +𝐾 ℓ(𝜕𝐶_𝑑/

2(𝜉)). Therefore,

ℓ_𝑑/

2(ℎ_𝑛(𝛼)) ≥ℓ(𝜓^∗

𝑛𝜌

1(𝛾)) −𝐾 ℓ(𝜕𝐶_𝑑/

2(𝜉))

which tends to∞as𝑛→ ∞, and moreover establishes the result of the lemma.

We are left to consider the case of an unbounded number of crossings into𝐶_𝑑/

2(𝜉) as we take𝑛 → ∞. The idea is that each time ℎ_𝑛(𝛼) crosses 𝜕𝐶_𝑑/

2(𝜉), there is a corresponding “down-crossing,” a curve that connects the point at which it exits to the new point of entry. If each down crossing is denoted by𝑐^𝑛

𝑗, then

∫

𝐴^𝑑/_𝑥²

𝜕 ℎ_𝑛(𝑥 , 𝑦

𝜕 𝑦

𝑑𝑦 ≥ ∑︁

𝑗

ℓ(𝑐^𝑛

𝑗).

Clearly, if the limit of these sums is infinite, then we are done. Hence, we assume the total length due to down-crossings is finite. This implies that there is an even integer𝐾 > 0 such that, as we take𝑛 → ∞, there are at most𝐾 down-crossing that exit𝐶_𝑑. Again using our chosen basepoint, let𝑏

1be the first crossing into𝐶_𝑑/

2(𝜉), and then let 𝑏

2 be the first crossing out of𝐶_𝑑/

2(𝜉) that exits𝐶_𝑑. Set 𝑏

3 to be the next entry point into𝐶_𝑑/

2(𝜉), and𝑏

4 the next exit point from 𝐶_𝑑(𝜉). We end up with𝐾_𝑛 ≤ 𝐾points𝑏

1, . . . 𝑏_𝐾

𝑛. Replace the segments ofℎ_𝑛(𝛼)between𝑏_𝑖and𝑏_𝑖+

1

with the correct arc on 𝜕𝐶_𝑑/

2(𝜉) that does not change the homotopy class, where 𝑖=1,3, . . . , 𝐾_𝑛−1 ≤ 𝐾−1. As above, we end up with a curve of length at most

ℓ_𝑑/

2(ℎ_𝑛(𝛼)) +𝐾 ℓ(𝜕𝐶_𝑑/

2(𝜉)) and homotopic to ℓ(𝜓^∗

𝑛𝜌

1(𝛾)). Using the minimizing property of geodesics once again, we have

ℓ_𝑑/

2(ℎ_𝑛(𝛼)) ≥ℓ(𝜓^∗

𝑛𝜌

1(𝛾)) −𝐾 ℓ(𝜕𝐶_𝑑/

2(𝜉)) → ∞,

and the resolution of this final case completes the proof. □ Returning to (4.15), this lemma showsE (𝜇_𝑛) → ∞, which is a contradiction. We can now conclude that each sequenceℓ(𝜓^∗

𝑛𝜌

1(𝛾))remains bounded above.

It is well-understood that the boundedness of the length spectrum implies that (𝜓^∗

𝑛𝜌

1) converges along a subsequence to some new Fuchsian representation 𝜌_∞ : Γ → PSL₂(R). We now restrict the [𝜇_𝑛],[𝜓_𝑛] to this chosen subsequence. Since the mapping class group acts properly discontinuously on the Teichmüller space, it follows that [𝜓^∗

𝑛𝜌

1] = [𝜌_∞] for all𝑛large enough. In particular, 𝜓_𝑛is isotopic to 𝜓_𝑚 for all𝑛, 𝑚large enough. This completes the proof of Lemma 4.3.3.

To finish the proof of Proposition 4.3.1, we know there is an𝑁such that[𝜓_𝑛] = [𝜓_𝑁] for all𝑛 ≥ 𝑁. As the mapping class group acts by isometries with respect to the metric𝑑_T,

𝑑_T( [𝜇_𝑛],[(𝜓⁻¹

𝑁 )^∗𝜇_∞]) =𝑑_T( [𝜓^∗

𝑁𝜇_𝑛],[𝜇_∞]) → 0, and hence [𝜇_𝑛]converges to [(𝜓⁻1

𝑁 )^∗𝜇_∞] as𝑛 → ∞. Proposition 4.3.1 is proved.

Uniqueness of critical points

The uniqueness statement in Theorem 4A amounts to showing that critical points ofEare unique. The main step in the uniqueness proof for closed surfaces [Tho17, Theorem 1] is a local computation ([Tho17, Lemma 2.4]) that goes through in our setting. For this reason, we omit the proof of uniqueness and invite the reader to see [Tho17]. The only real difference is that we must address convergence of various integrals, and use our Lemma 4.2.14 instead of the usual energy minimizing property. We’ve shown this type of calculation throughout the chapter, so we feel comfortable leaving it to the interested reader.

The mapsΨ^𝜃

We now begin the proof of Theorem 4B. Generalizing the map Ψ from [Tho17, subsection 2.3]), we define the map

Ψ^𝜃 :T (Γ) ×Repc(Γ, 𝐺) →ASDc(Γ, 𝐺) ⊂ T_c(Γ) ×Repc(Γ, 𝐺)

as follows. When chas no hyperbolic classes, we implicitly assume there is no 𝜃. Begin with a hyperbolic surface (Σ, 𝜇) and a reductive representation 𝜌

2 : Γ → PSL(2,R). Associated to this representation, we take a 𝜌

2-equivariant harmonic map with twist parameter𝜃, 𝑓^𝜃 : (Σ˜,𝜇˜) → (𝑋 , 𝜈). We proved in [Sag19, Theorem 1.4] that there exists a unique Fuchsian representation 𝜌

1 : Γ → PSL(2,R) and a unique 𝜌

1-equivariant harmonic diffeomorphismℎ^𝜃 : (Σ˜, 𝜇) → (H, 𝜎) that has Hopf differential Φ(ℎ^𝜃) = Φ(𝑓^𝜃). From [Sag19, Proposition 3.13], the mapping 𝐹 =(ℎ^𝜃, 𝑓^𝜃)is a spacelike maximal immersion into the pseudo-Riemannian product, and hence 𝜌

1almost strictly dominates𝜌

2. Ψ^𝜃 is defined by Ψ^𝜃( [𝜇],[𝜌

2])= ( [𝜌

1],[𝜌

2]).

It is clear that the map is fiberwise in the sense of the statement of Theorem 4B.

Theorem 4A showsΨ^𝜃 is bijection, and the content of Theorem 4B is that Ψ^𝜃 is a homeomorphism.

Remark 4.3.5. We do not know ifΨ^𝜃 = Ψ^𝜃′ for𝜃 ≠ 𝜃^′! Proof of Theorem 4B: continuity

From now on we refine our notation: when the representation 𝜌 is not implicit, the 𝜌-equivariant harmonic map for a metric 𝜇 with zero twist parameter will be denoted 𝑓

𝜌

𝜇, unless specified otherwise.

The proof of continuity is almost identical for each twist parameter, so we work only withΨ = Ψ⁰. Let us also assume for the remainder of this section that there is a single peripheral 𝜁. Lifting various equivalence relations, Ψis described by a composition

(𝜇, 𝜌

2) ↦→ (𝑓

𝜌2

𝜇 , 𝜌

2) ↦→ (Φ(𝑓

𝜌2

𝜇 ), 𝜌

2) ↦→ (𝜌

1, 𝜌

2), where 𝜌

1 is the holonomy of the hyperbolic metric 𝜇^′ on Σ such that the associ- ated harmonic map from (Σ, 𝜇) → (Σ, 𝜇^′) has Hopf differential Φ(𝑓

𝜌2

𝜇 ). [Sag19, Theorem 1.4] implies the association fromΦ(𝑓

𝜌2

𝜇 ) ↦→ 𝜌

1is continuous. Here, the topology on the space of holomorphic quadratic differentials is the Fréchet topology

coming from taking 𝐿¹-norms over a compact exhaustion. To show continuity of 𝜇↦→ 𝑓

𝜌2

𝜇 ↦→Φ(𝑓

𝜌2

𝜇 )with respect to this topology, note that, by the constructions of Section 4.2, we already have continuity in the Teichmüller coordinate. Continuity will thus follow from the results below.

Lemma 4.3.6. Suppose representations (𝜌_𝑛)^∞

𝑛=1 converge to an irreducible 𝜌 in Hom(Γ, 𝐺), and all such representations project to Rep_c(Γ, 𝐺). Then 𝑓

𝜌𝑛

𝜇 converges to 𝑓

𝜌

𝜇 in the𝐶^∞sense on compacta as𝑛→ ∞.

Below, we work in fixed local sections of the principal bundles 𝜒_c(Γ, 𝐺) → Repc(Γ, 𝐺) 𝜒_c(Γ,PSL(2,R)) → Repc(Γ,PSL(2,R)) (having assumed in the in- troduction that such things exist). We choose a basepoint for the𝜋

1so that we can view these local systems through their holonomies, which are genuine representa- tions.

Proof. We may assume each 𝜌_𝑛 is irreducible and that there is a single cusp. The parabolic and elliptic cases are much simpler than the hyperbolic case, so we treat only the latter. Set 𝑓_𝑛 = 𝑓

𝜌𝑛

𝜇 , 𝑓 = 𝑓

𝜌

𝜇 and let 𝜑_𝑛, 𝜑 be the approximation maps for 𝑓

𝜌𝑛

𝜇 , 𝑓

𝜌

𝜇 from 3.1. These maps project a cusp neighbourhood onto a geodesic, but now the geodesic is varying with 𝜌_𝑛, and converging in the Gromov-Hausdorff sense to the geodesic for𝜌. If the𝜑_𝑛’s can be chosen to vary smoothly, then we have a uniform bound on the total energies (recall they depend on a choice of a constant speed map onto their geodesic). As discussed in Section 4.2 (in particular the proof of Lemma 4.2.11), this yields a uniform total energy bound for 𝑓_𝑛on compacta. Via the argument described early in Section 4.2, the maps 𝑓_𝑛𝐶^∞-converge on compacta along a subsequence to some limiting map 𝑓^∞. The mapping is necessarily 𝜌- equivariant and harmonic. To see that 𝑓^∞ = 𝑓, recall that 𝑓 is characterized by the fact that its Hopf differential has a pole of order 2 at the cusp and the residue has a specific complex argument. To check this property, from the uniqueness in [Sag19, Theorem 1.1] it suffices to check 𝑑(𝑓_∞, 𝜑) < ∞, and the standard contradiction argument will also show there is no need to pass to a subsequence. As remarked previously, the proof of [Sag19, Lemma 5.2] shows𝑑(𝑓_𝑛, 𝜑_𝑛)is maximized on𝜕Σ₂. If𝑛_𝑘 is the subsequence, then taking 𝑘 → ∞, via compactness we do win

𝑑(𝑓_∞, 𝜑_∞) ≤lim sup

𝑘→∞ max

𝜕Σ₂

𝑑(𝑓_𝑛

𝑘, 𝜑_𝑛

𝑘) ≤max

𝜕Σ₂

𝑑(𝑓_∞, 𝜑_∞) +1< ∞.

We are left to argue that one can choose the𝜑_𝑛so that𝜑_𝑛 →𝜑in the𝐶^∞sense. We realize𝜌, 𝜌_𝑛as monodromies of flat connections∇,∇𝑛respectively on an𝑋-bundle

𝐸 over Σ with structure group 𝐺. This is possible when all 𝜌_𝑛 lie in the same component of the relative representation space ([Lab13, Corollay 6.1.2] generalizes to relative representation spaces), which we are free to assume. The pullback bundle with respect to the universal covering ˜Σ → Σidentifies with the trivial bundle and also pulls∇,∇𝑛back to flat connections ˜∇,∇˜𝑛. That is, up to an isomorphism, we have a family of commutative diagrams

(𝑋 ×Σ˜,∇˜𝑛) (𝐸 ,∇𝑛)

Σ˜ Σ.

The bundle𝐸 and connections∇𝑛are constructed by choosing a good covering of Σ and depend on the local section of the principal bundle 𝜒_c(Γ, 𝐺) →Repc(Γ, 𝐺) (see the proof of [Lab13, Lemma 6.1.1]). We can build a good covering by glueing a good covering of a relatively compact tubular neighbourhood of(Σ₂, 𝜇)with a good covering of a tubular neighbourhood of (Σ\Σ₂, 𝜇). This ensures a lower bound on the𝜇-radius insideΣ₂(note that we cannot do this on all ofΣ). With this constraint, the local systems can be prescribed so that ∇𝑛 → ∇in the 𝐶^∞ sense on Σ₂ in the affine space of connections.

The map 𝜑 induces a section 𝑠 of the bundle (𝐸 ,∇), which can also be seen as a section 𝑠_𝑛 of (𝐸 ,∇𝑛). With respect to the diagram above, 𝑠_𝑛 pulls back to a 𝜌_𝑛-equivariant map ˜𝜑_𝑛 :H → 𝑋, and by our comments above, ˜𝜑_𝑛 → 𝜑in the𝐶^∞ sense. The maps ˜𝜑_𝑛 are most likely not harmonic and may not project onto the geodesic axis of 𝜌_𝑛(𝜁). However, because they are converging to 𝜑, the geodesic curvature of ˜𝜑_𝑛(𝜕Σ˜₂) is tending to 0. Moreover,𝜑|_𝜕Σ˜

2 is arbitrarily close to some parametrization of a geodesic. So, we set𝜑_𝑛to be the harmonic map with boundary data equal to this nearby geodesic projection. From energy control on ˜𝜑_𝑛, we get the same for 𝜑_𝑛, and hence convergence along a subsequence to a harmonic map 𝜑_∞. From the boundary values, we must have 𝜑_∞ = 𝜑, and as usual we can show

there is no need to pass to a subsequence. □

For a reducible representation, the harmonic maps are not unique but differ by translations along a geodesic. However, the Hopf differential is unique. We can choose a sequence𝜑_𝑛 → 𝜑as in the previous proposition—boundary values are fixed so we do have uniqueness for these harmonic maps. Then we take the harmonic maps 𝑓_𝑛 built from using the approximating maps𝜑_𝑛, and the proof above goes through,

with the minor modification that we must rescale the approximating maps as in [Sag19, Proposition 5.1]. 𝐶^∞ convergence implies the Hopf differentials converge.

Remark 4.3.7. For different twist parameters, we simply precompose every𝜑_𝑛with a fractional Dehn twist. Using (4.12), the energies are uniformly controlled.

Asymmetric metrics on Teichmüller spaces

In [GK17, Section 8], Guéritaud-Kassel define asymmetric pseudo-metrics on de- formation spaces of geometrically finite hyperbolic manifolds. These metrics are natural generalizations of Thurston’s asymmetric metric on Teichmüller space. We will use such a metric in the proof of bi-continuity ofΨ^𝜃.

For[𝜌

1],[𝜌

2] ∈ T_𝔠(𝑆_𝑔,𝑛), we set 𝐶(𝜌

1, 𝜌

2) =inf{Lip(𝑓) : 𝑓 :H→His(𝜌

1, 𝜌

2) −equivariant}. The metric𝑑_{𝑇 ℎ} :T_𝔠(𝑆_𝑔,𝑛) × T_𝔠(𝑆_𝑔,𝑛) → Ris defined by

𝑑_{𝑇 ℎ}( [𝜌

1],[𝜌

2]) =log

𝐶(𝜌

1, 𝜌

2)𝛿(𝜌

1) 𝛿(𝜌

2)

. Here𝛿(𝜌

1)is the critical exponent of 𝜌

1: 𝛿(𝜌

1) = lim

𝑅→∞

1

𝑅log #(𝑗(Γ) ·𝑝∩𝐵_𝑝(𝑅)),

where 𝑝is any point inHand𝐵_𝑝(𝑅) is the ball of radius𝑟 centered at 𝑝. Alterna- tively, 𝛿(𝜌

1) is the Hausdorff dimension of the limit set of 𝜌

1in 𝜕_∞H = 𝑆¹. If all 𝑎_𝑗 =0, so that all critical exponents are 1 andT_𝔠(𝑆_𝑔,𝑛)is the ordinary Teichmüller space of a surface of finite volume, then this agrees with Thurston’s asymmetric metric introduced in [Thu98]. Here asymmetric means that in general,

𝑑_{𝑇 ℎ}( [𝜌

1],[𝜌

2]) ≠ 𝑑_{𝑇 ℎ}( [𝜌

2],[𝜌

1]).

Proposition 4.3.8(Guéritaud-Kassel, Lemma 8.1 in [GK17]). The function 𝑑_{𝑇 ℎ} : T_c(Γ) × T_c(Γ) → Ris a continuous asymmetric metric.

Remark 4.3.9. The correction factor𝛿(𝜌

1)/𝛿(𝜌

2)is needed for non-negativity. For a general hyperbolic𝑛-manifold, continuity may fail and the generalization may be just an asymmetric pseudo-metric (𝑑(𝑥 , 𝑦) need not imply𝑥 =𝑦).

This metric allows us to control translation lengths nicely. Consider a left open ball around𝜌

2:

𝐵_{𝑇 ℎ}(𝜌

2, 𝐶) ={[𝜌

1] ∈ T_c(Γ) : 𝑑_{𝑇 ℎ}(𝜌

1, 𝜌

2) < 𝐶}.