Angel Surfaces: Planar Maps & Minimal Surfaces

• Angel surfaces are defined in two ways: as Riemann surfaces from random planar maps and as complete minimal surfaces in ℝ³ with two ends and minimal total curvature.
• In the random planar map approach, equilateral triangles are glued to yield a parabolic surface where Brownian motion is recurrent.
• In minimal surface theory, the construction produces surfaces of genus p with one catenoidal and one Enneper end, attaining a total absolute curvature of 4π(p+1).

“Angel surfaces” denotes two mathematically distinct constructions that share a name but arise in different research programs. In random planar map theory, the term refers to the Riemann surfaces obtained by gluing equilateral triangles according to the Angel–Schramm Uniform Infinite Planar Triangulation or Sheffield’s infinite necklace, and the central result is that these surfaces are parabolic, equivalently that Brownian motion on them is recurrent (Gill et al., 2011). In minimal-surface theory, an Angel surface is a complete minimal immersion of genus $p\ge 1$ with exactly two ends, one catenoidal and one Enneper-type, realizing the least total absolute curvature $4\pi(p+1)$; higher-genus examples exist for every integer $p\ge 1$ (Bardhan et al., 4 Sep 2025). The common label therefore covers both a conformal-type problem for random planar maps and an existence problem for complete minimal surfaces.

1. Terminological scope and mathematical settings

The two uses of the term can be summarized as follows.

Usage	Ambient setting	Defining feature
Angel–Schramm / infinite-necklace surfaces	Random planar maps and Riemann surfaces	Equilateral-triangle gluing yields a parabolic surface
Angel surfaces in $\mathbb{R}^3$	Complete minimal-surface theory	Genus $p$, two ends, least total absolute curvature

In the first sense, a triangulation $T$ is a locally finite simplicial $2$-complex whose underlying graph is embedded in $S^2$ so that each face has exactly three boundary edges. One forms a length-metric space $M(T)$ by declaring each combinatorial triangle to be a Euclidean equilateral triangle of side length $1$ and gluing them isometrically along edges according to the combinatorics of $4\pi(p+1)$0. When $4\pi(p+1)$1 is infinite and one-ended, the resulting topological space is simply connected and non-compact; by the uniformization theorem it carries a unique conformal structure, denoted $4\pi(p+1)$2 (Gill et al., 2011).

In the second sense, an Angel surface of genus $4\pi(p+1)$3 is a complete, conformally immersed minimal surface

$4\pi(p+1)$4

for which $4\pi(p+1)$5 is a compact Riemann surface of genus $4\pi(p+1)$6 minus two points, with exactly two ends, one asymptotic to a catenoid and the other asymptotic to an Enneper surface. Its total absolute Gaussian curvature satisfies Osserman’s formula

$4\pi(p+1)$7

and this is the least absolute curvature among complete minimal surfaces of genus $4\pi(p+1)$8 with two ends (Bardhan et al., 4 Sep 2025).

A recurrent source of confusion is the assumption that the phrase has a single standard meaning. The literature represented here instead uses it for two independent objects: one governed by conformal type and Brownian recurrence, the other by Weierstrass data, period conditions, and least-curvature existence.

2. Angel surfaces from random planar maps

The random-surface construction begins with infinite triangulations. The Angel–Schramm Uniform Infinite Planar Triangulation is the distributional local limit of uniformly rooted finite triangulations of $4\pi(p+1)$9 as the number of vertices tends to infinity. Equivalently, it can be obtained by repeatedly gluing triangles in an “unbiased” way so that each new triangle is equally likely to appear at any existing face (Gill et al., 2011).

A second model is Sheffield’s infinite necklace construction. One alternately grows a finite triangulated disc $p\ge 1$0 in the upper half-plane by gluing one new triangle along a distinguished active edge with colored endpoints Blue and Red. At each step there are four choices—$p\ge 1$1—corresponding to adding a new blue or red vertex or gluing to an existing boundary neighbor. If $p\ge 1$2 is an infinite i.i.d. sequence with $p\ge 1$3, then $p\ge 1$4 is a rooted disc triangulation with $p\ge 1$5 faces. An independent i.i.d. sequence $p\ge 1$6 similarly produces a lower-half-plane triangulation. Gluing the two half-plane triangulations $p\ge 1$7 and $p\ge 1$8 along their boundary yields the infinite necklace surface; almost surely the result is one-ended and simply connected (Gill et al., 2011).

The passage from combinatorics to complex analysis is canonical. The metric space $p\ge 1$9 is covered by local charts of three kinds: the interior of a triangle, the union of two adjacent triangles along an edge, and the union of the $\mathbb{R}^3$0 triangles incident to a vertex of degree $\mathbb{R}^3$1. In the vertex chart, one uses

$\mathbb{R}^3$2

which sends the total cone angle $\mathbb{R}^3$3 to $\mathbb{R}^3$4. Transition maps are analytic, so $\mathbb{R}^3$5 acquires a Riemann-surface structure $\mathbb{R}^3$6 (Gill et al., 2011).

This construction isolates the geometric content of the underlying random map. The resulting question is not merely graph-theoretic recurrence but the conformal type of the glued surface itself.

3. Uniformization, parabolicity, and recurrence

Because $\mathbb{R}^3$7 is simply connected and non-compact, the Riemann mapping theorem yields either a conformal homeomorphism

$\mathbb{R}^3$8

or

$\mathbb{R}^3$9

These correspond respectively to the parabolic and hyperbolic cases, unique up to post-composition by an affine map in the first case or a Möbius automorphism of $p$0 in the second (Gill et al., 2011).

For simply connected Riemann surfaces, parabolicity has several equivalent formulations. A surface is parabolic if it is conformally $p$1, equivalently if it admits no nonconstant bounded harmonic functions, equivalently if Brownian motion is recurrent. It is hyperbolic if it is conformally $p$2, admits nonzero bounded harmonic functions, and Brownian motion is transient. The Ahlfors–Beurling criterion expresses the same dichotomy in terms of capacity: $p$3

The main theorem concerns subsequential local limits of unbiased finite disc triangulations. If $p$4 is a sequence of random finite unbiased disc-triangulations, each rooted at a uniformly chosen oriented face, and if $p$5 converges in distribution to an infinite one-ended triangulation $p$6 while the graph distance $p$7 in law, then the random Riemann surface $p$8 is almost surely parabolic. The Angel–Schramm UIPT and Sheffield’s infinite necklace satisfy these hypotheses, hence their associated surfaces are almost surely parabolic (Gill et al., 2011).

The proof combines uniformization, compactness, and a Benjamini–Schramm argument on embeddings. For each finite triangulation one normalizes the uniformizing map so that the center $p$9 of the root face maps to $T$0 and the next-closest interstice has distance $T$1. Koebe distortion and Montel yield tightness. A limit-point lemma then shows that in an unbiased distributional limit the random set of embedded face centers has at most one finite accumulation point almost surely. Bounded geometry of interstices implies that a hyperbolic embedding into $T$2 would force accumulation everywhere on the boundary, contradicting the single-accumulation-point conclusion. Hence the limit surface must be conformally $T$3 (Gill et al., 2011).

A common misconception is to read this result as a purely combinatorial recurrence theorem. The statement is stronger in a different direction: it concerns Brownian motion on the glued-triangle Riemann surface, and recurrence follows from parabolic conformal type.

4. Angel surfaces as complete minimal surfaces

In minimal-surface theory, Angel surfaces are defined by topology, asymptotics, and curvature. For genus $T$4, an Angel surface is a complete minimal immersion $T$5 where $T$6 is a compact Riemann surface of genus $T$7 minus two points, the two punctures giving exactly two ends: one catenoidal and one Enneper-type. The total absolute Gaussian curvature is

$T$8

and this equals the lower bound for complete minimal surfaces of genus $T$9 with two ends (Bardhan et al., 4 Sep 2025).

The existence theorem states that for every integer $2$0 there exists such a complete minimal immersion with exactly these two ends and total absolute curvature $2$1. For genus zero the unique example is the catenoid; for genus one the first Angel surface with one catenoid end and one Enneper end was built by Fujimori–Shoda (Bardhan et al., 4 Sep 2025).

The least-curvature property is derived from the classical theorem of Chern–Osserman and Osserman’s inequality: any complete minimal surface of genus $2$2 with $2$3 ends satisfies

$2$4

For $2$5, the lower bound is $2$6, exactly the value achieved by Angel surfaces. In this sense the class is extremal within the two-ended genus-$2$7 problem.

This terminology is specific to the minimal-surface literature. Here the essential issue is not conformal type in the plane-versus-disc sense, but the realization of prescribed end behavior and sharp curvature under the Weierstrass–Enneper representation.

5. Weierstrass–Enneper data and the orthodisk construction

The minimal immersion is encoded by meromorphic data. For $2$8, any conformal minimal immersion is given by a meromorphic function $2$9 and a holomorphic one-form $S^2$0 such that

$S^2$1

and for every $S^2$2,

$S^2$3

Then

$S^2$4

defines a well-defined immersion in $S^2$5 (Bardhan et al., 4 Sep 2025).

For Angel surfaces of genus $S^2$6, the explicit data are placed on the hyperelliptic curve

$S^2$7

where

$S^2$8

punctured at $S^2$9 and $M(T)$0. One takes

$M(T)$1

with $M(T)$2 real. Near $M(T)$3, $M(T)$4 and $M(T)$5, giving a catenoidal end; near $M(T)$6, $M(T)$7 and $M(T)$8, giving an Enneper end. The total curvature is

$M(T)$9

The period problem is reformulated by the Weber–Wolf orthodisk method. A generalized orthodisk is specified by $1$0, with a Schwarz–Christoffel map

$1$1

mapping the upper half-plane to a Euclidean polygon whose vertex $1$2 has interior angle $1$3. Two orthodisks $1$4 and $1$5 are chosen so that the exponents $1$6 control the zero/pole orders of $1$7, the exponents $1$8 control those of $1$9, and $4\pi(p+1)$00 ensures that $4\pi(p+1)$01 is a globally well-defined quadratic differential (Bardhan et al., 4 Sep 2025).

Because the two ends are of different type, the flat structures are not fully symmetric. Instead the construction introduces “partial symmetry”: a staircase of orthogonal edges of lengths $4\pi(p+1)$02, in two types distinguished by whether they start with a horizontal or vertical step, is inserted between two classical genus-$4\pi(p+1)$03 polygons. After rotation by $4\pi(p+1)$04, the $4\pi(p+1)$05th horizontal step of one polygon becomes the complex conjugate of the $4\pi(p+1)$06th vertical step of the other, and vice versa. This is the key geometric device that replaces full reflexive symmetry in higher genus.

6. Existence proof, generalizations, and conceptual distinctions

The existence theorem for higher-genus Angel surfaces is obtained through a height-function argument on the parameter space of partially symmetric orthodisk pairs. For a fixed genus-$4\pi(p+1)$07 base configuration $4\pi(p+1)$08, the space $4\pi(p+1)$09 of all e-conjugate, partially symmetric orthodisk pairs is parametrized by the $4\pi(p+1)$10 staircase lengths $4\pi(p+1)$11. The real height function

$4\pi(p+1)$12

is defined as a sum of squared differences involving $4\pi(p+1)$13, $4\pi(p+1)$14, and their reciprocals over standard loops $4\pi(p+1)$15 encircling adjacent punctures. A zero of $4\pi(p+1)$16 forces equality of the relevant extremal lengths, and a Schwarz reflection argument then makes the two orthodisks conformally identical, that is, reflexive. Properness implies $4\pi(p+1)$17 at the boundary of parameter space, so a minimum with $4\pi(p+1)$18 occurs in the interior, yielding a solution of the period problem and hence an Angel surface of genus $4\pi(p+1)$19. The inductive step starts from Fujimori–Shoda’s genus-$4\pi(p+1)$20 solution and adds two new conjugate stair-steps to pass from genus $4\pi(p+1)$21 to genus $4\pi(p+1)$22 (Bardhan et al., 4 Sep 2025).

The random-surface theory has a different generalization. The parabolicity proof extends to any subsequential distributional limit of finite unbiased disc-triangulations, or more generally $4\pi(p+1)$23-angulations with uniformly bounded face-degrees, provided one-endedness and $4\pi(p+1)$24 hold. The implication is that many other random planar-map models arising as local limits admit recurrent Brownian motion on their glued-triangle surfaces (Gill et al., 2011).

Taken together, the two theories illustrate two unrelated roles played by Riemann surfaces in current research. In the random planar-map setting, the surface is obtained from a combinatorial gluing and then classified by uniformization and recurrence. In the minimal-surface setting, the Riemann surface is the conformal domain on which meromorphic data are prescribed in order to produce a complete immersion in $4\pi(p+1)$25. The shared nomenclature should therefore be read as a terminological coincidence rather than as an indication of a common construction.