De l’inf´erence g´eom´etrique
au transport optimal num´erique
Quentin M´ erigot
Laboratoire Jean Kuntzmann, CNRS / Universit´e de Grenoble
Soutenance d’habilitation
Overview
1. Geometric inference from noisy data 2. Computational optimal transport
3. Far-Field Reflector Problem
4. Discretization of Functionals Involving the Monge-Amp` ere Operator.
1. Geometric inference from noisy data
PhD: Geometric Inference from Point Clouds
dK = distance function to K pK = projection on K
K ⊆ Rd
PhD: Geometric Inference from Point Clouds
dK = distance function to K pK = projection on K
K ⊆ Rd
KR = d−1K ([0, R])
PhD: Geometric Inference from Point Clouds
µK,R = pK# Hd
Kr dK = distance function to K
pK = projection on K K ⊆ Rd
KR = d−1K ([0, R])
Boundary measure:
PhD: Geometric Inference from Point Clouds
I Estimation of Federer’s curvature measures from Hausdorff approximation.
[Chazal, Cohen-Steiner, M. 2008]
µK,R = pK# Hd
Kr dK = distance function to K
pK = projection on K K ⊆ Rd
KR = d−1K ([0, R])
Boundary measure:
P =
PhD: Geometric Inference from Point Clouds
I Estimation of Federer’s curvature measures from Hausdorff approximation.
[Chazal, Cohen-Steiner, M. 2008]
µK,R = pK# Hd
Kr dK = distance function to K
pK = projection on K K ⊆ Rd
KR = d−1K ([0, R])
Boundary measure:
P =
PhD: Geometric Inference from Point Clouds
I Anisotropic version: Voronoi covariance measures
I Estimation of Federer’s curvature measures from Hausdorff approximation.
[Chazal, Cohen-Steiner, M. 2008]
[M., Ovsjanikov, Guibas 2009]
I Handling outliers −→ distance to a probability measure [Chazal, Cohen-Steiner, M. 2010]
µK,R = pK# Hd
Kr dK = distance function to K
pK = projection on K K ⊆ Rd
KR = d−1K ([0, R])
Boundary measure:
P =
Simplifying the Distance to a Measure
d2P,k(x) = minpi6=...6=pk∈P k1 Pk
i=1 kx − pik2 k-Distance: Let P ⊆ Rd, k ∈ {1, . . . , |P|}
Simplifying the Distance to a Measure
d2P,k(x) = minpi6=...6=pk∈P k1 Pk
i=1 kx − pik2
d2P,k(x) = minb∈Baryk
P kx − bk2 + wb k-Distance: Let P ⊆ Rd, k ∈ {1, . . . , |P|}
Simplifying the Distance to a Measure
d2P,k(x) = minpi6=...6=pk∈P k1 Pk
i=1 kx − pik2
d2P,k(x) = minb∈Baryk
P kx − bk2 + wb
−→ 1-semiconcave k-Distance: Let P ⊆ Rd, k ∈ {1, . . . , |P|}
−→ sublevel sets = ∪ balls
−→ exponential # of barycenters relation to weighted Voronoi diagrams
Simplifying the Distance to a Measure
d2P,k(x) = minpi6=...6=pk∈P k1 Pk
i=1 kx − pik2
d2P,k(x) = minb∈Baryk
P kx − bk2 + wb k-Distance: Let P ⊆ Rd, k ∈ {1, . . . , |P|}
Simplification problem: ∃ ”small” B ⊆ BarykP s.t. k dP,k −dBk∞ is ”small” ? d2B(x) = minb∈B kx − bk2 + wb
Simplifying the Distance to a Measure
d2P,k(x) = minpi6=...6=pk∈P k1 Pk
i=1 kx − pik2
d2P,k(x) = minb∈Baryk
P kx − bk2 + wb k-Distance: Let P ⊆ Rd, k ∈ {1, . . . , |P|}
Simplification problem: ∃ ”small” B ⊆ BarykP s.t. k dP,k −dBk∞ is ”small” ? d2B(x) = minb∈B kx − bk2 + wb
∀x ∈ K, ∀r ≤ diam(K), µK(B(x, r)) ≥ αKr` Ex: Area measure on a compact surface → ` = 2.
K := spt(µ ) x
B(x, r)
Assume µK is supported on K and ∃αK > 0 such that I Positive result:
[Guibas, M., Morozov, 2011]
Simplifying the Distance to a Measure
d2P,k(x) = minpi6=...6=pk∈P k1 Pk
i=1 kx − pik2
d2P,k(x) = minb∈Baryk
P kx − bk2 + wb k-Distance: Let P ⊆ Rd, k ∈ {1, . . . , |P|}
Simplification problem: ∃ ”small” B ⊆ BarykP s.t. k dP,k −dBk∞ is ”small” ? d2B(x) = minb∈B kx − bk2 + wb
∀x ∈ K, ∀r ≤ diam(K), µK(B(x, r)) ≥ αKr` Assume µK is supported on K and ∃αK > 0 such that I Positive result:
K µK
P −→ if W2(µP , µK) is small, then one can construct a good approximation of dP,k involving only |P| barycenters.
”Witnessed k-distance”
[Guibas, M., Morozov, 2011]
Simplifying the Distance to a Measure
d2P,k(x) = minpi6=...6=pk∈P k1 Pk
i=1 kx − pik2
d2P,k(x) = minb∈Baryk
P kx − bk2 + wb k-Distance: Let P ⊆ Rd, k ∈ {1, . . . , |P|}
Simplification problem: ∃ ”small” B ⊆ BarykP s.t. k dP,k −dBk∞ is ”small” ? d2B(x) = minb∈B kx − bk2 + wb
∀x ∈ K, ∀r ≤ diam(K), µK(B(x, r)) ≥ αKr` Assume µK is supported on K and ∃αK > 0 such that I Positive result:
µK
P −→ if W2(µP , µK) is small, then one can construct a good approximation of dP,k involving only |P| barycenters.
”Witnessed k-distance”
−→ Application to reconstruction and topological data analysis.
[Guibas, M., Morozov, 2011]
Simplifying the Distance to a Measure
d2P,k(x) = minpi6=...6=pk∈P k1 Pk
i=1 kx − pik2
d2P,k(x) = minb∈Baryk
P kx − bk2 + wb k-Distance: Let P ⊆ Rd, k ∈ {1, . . . , |P|}
Simplification problem: ∃ ”small” B ⊆ BarykP s.t. k dP,k −dBk∞ is ”small” ? d2B(x) = minb∈B kx − bk2 + wb
∀x ∈ K, ∀r ≤ diam(K), µK(B(x, r)) ≥ αKr` Assume µK is supported on K and ∃αK > 0 such that I Positive result:
I Negative result: exponential lower bound when P is drawn randomly from Sd−1. K
µK
P −→ if W2(µP , µK) is small, then one can construct a good approximation of dP,k involving only |P| barycenters.
”Witnessed k-distance”
−→ Application to reconstruction and topological data analysis.
[Guibas, M., Morozov, 2011]
6
Sharp Feature Estimation
Input: 50k point set P
I Application of the witnessed k-distance to curvature estimation and edge recovery
6
Sharp Feature Estimation
Input: 50k point set P Output: estimated edges and directions I Application of the witnessed k-distance to curvature estimation and edge recovery
[Cuel, Lachaud, M., Thibert 2014]
6
Sharp Feature Estimation
Input: 50k point set P Output: estimated edges and directions I Application of the witnessed k-distance to curvature estimation and edge recovery
[Cuel, Lachaud, M., Thibert 2014]
From curvature estimation to the inverse problem: Monge-Amp`ere equations.
2. Computational optimal transport
Computational optimal transport
Hungarian algorithm linear programming
Discrete source and target
Bertsekas’ auction algorithm
αi βj
Computational optimal transport
Hungarian algorithm linear programming
Discrete source and target
Bertsekas’ auction algorithm
αi βj
Source and target with density:
Benamou-Brenier ’00 Loeper-Rapetti ’05
Benamou-Froese-Oberman ’12
Computational optimal transport
Hungarian algorithm linear programming
Discrete source and target
Bertsekas’ auction algorithm
αi βj
Source with density, discrete target:
Kitagawa ’12
Source and target with density:
Benamou-Brenier ’00 Loeper-Rapetti ’05
Benamou-Froese-Oberman ’12
Theoretical use: Alexandrov, Pogorelov, etc.
Aurenhammer, Hoffmann, Aronov ’98 Cullen ’89
Computational optimal transport
Hungarian algorithm linear programming
Discrete source and target
Bertsekas’ auction algorithm
αi βj
Source with density, discrete target:
Kitagawa ’12
Source and target with density:
Benamou-Brenier ’00 Loeper-Rapetti ’05
Benamou-Froese-Oberman ’12
Flexibility for the cost function but computationally expensive
Theoretical use: Alexandrov, Pogorelov, etc.
Aurenhammer, Hoffmann, Aronov ’98 Cullen ’89
Computational optimal transport
Hungarian algorithm linear programming
Discrete source and target
Bertsekas’ auction algorithm
αi βj
Source with density, discrete target:
Kitagawa ’12
Source and target with density:
Benamou-Brenier ’00 Loeper-Rapetti ’05
Benamou-Froese-Oberman ’12
Theoretical use: Alexandrov, Pogorelov, etc.
Aurenhammer, Hoffmann, Aronov ’98 Cullen ’89
Optimal transport: Monge problem
ρ ∈ Pac(X) X = d-manifold ν = P
y∈Y νyδy ∈ P(Y ) Y finite
y
prob. measure with density
X
Y
prob. measure
Optimal transport: Monge problem
ρ ∈ Pac(X) X = d-manifold ν = P
y∈Y νyδy ∈ P(Y ) Y finite
T−1(y) y
Transport map: T#ρ = ν iff
∀y ∈ Y, ρ(T−1({y})) = νy
T : X → Y
X
Y
Optimal transport: Monge problem
ρ ∈ Pac(X) X = d-manifold ν = P
y∈Y νyδy ∈ P(Y ) Y finite
T−1(y) y
Transport map: T#ρ = ν iff
∀y ∈ Y, ρ(T−1({y})) = νy
Cost function: c : X × Y → R T : X → Y
X
Y
Optimal transport: Monge problem
ρ ∈ Pac(X) X = d-manifold ν = P
y∈Y νyδy ∈ P(Y ) Y finite
Monge problem: Tc(ρ, ν) := min{R
c(x, T(x)) d ρ(x); T#ρ = ν}
T−1(y) y
Transport map: T#ρ = ν iff
∀y ∈ Y, ρ(T−1({y})) = νy
Cost function: c : X × Y → R T : X → Y
X
Y
Optimal transport: Monge problem
ρ ∈ Pac(X) X = d-manifold ν = P
y∈Y νyδy ∈ P(Y ) Y finite
Monge problem: Tc(ρ, ν) := min{R
X c(x, T(x)) d ρ(x); T#ρ = ν}
T−1(y) y
Transport map: T#ρ = ν iff
∀y ∈ Y, ρ(T−1({y})) = νy
Cost function: c : X × Y → R T : X → Y
X
Y
ex: X, Y ⊆ Rd, c = 12k.k2 Wasserstein distance W2 = Tc1/2
Weighted Voronoi and Optimal Transport
We assume (Twist): ∀x ∈ X, the map y ∈ Y 7→ ∇xc(x, y) is injective.
Weighted Voronoi and Optimal Transport
Generalized Voronoi diagram: Given ψ : Y → R,
We assume (Twist): ∀x ∈ X, the map y ∈ Y 7→ ∇xc(x, y) is injective.
Vorψc (y) = {x ∈ X; ∀z ∈ Y, c(x, y) + ψ(y) ≤ c(x, z) + ψ(z)}
y z
Weighted Voronoi and Optimal Transport
Generalized Voronoi diagram: Given ψ : Y → R,
We assume (Twist): ∀x ∈ X, the map y ∈ Y 7→ ∇xc(x, y) is injective.
Vorψc (y) = {x ∈ X; ∀z ∈ Y, c(x, y) + ψ(y) ≤ c(x, z) + ψ(z)}
y
z I the Voronoi cells partition X up to a negligible set
Weighted Voronoi and Optimal Transport
Generalized Voronoi diagram: Given ψ : Y → R,
Tcψ(x) = arg miny∈Y c(x, y) + ψ(y)
We assume (Twist): ∀x ∈ X, the map y ∈ Y 7→ ∇xc(x, y) is injective.
Vorψc (y) = {x ∈ X; ∀z ∈ Y, c(x, y) + ψ(y) ≤ c(x, z) + ψ(z)}
y z
is uniquely defined Hd-almost everywhere
I the Voronoi cells partition X up to a negligible set I the ”generalized nearest neighbor” map
Weighted Voronoi and Optimal Transport
Generalized Voronoi diagram: Given ψ : Y → R,
Tcψ(x) = arg miny∈Y c(x, y) + ψ(y)
We assume (Twist): ∀x ∈ X, the map y ∈ Y 7→ ∇xc(x, y) is injective.
Vorψc (y) = {x ∈ X; ∀z ∈ Y, c(x, y) + ψ(y) ≤ c(x, z) + ψ(z)}
y z
is uniquely defined Hd-almost everywhere
Lemma: Given ψ : Y → the map Tψ is a c-optimal transport map between I the Voronoi cells partition X up to a negligible set
I the ”generalized nearest neighbor” map
Optimal transport as concave maximization
Theorem: Finding a c-O.T between ρ with density and ν = P
Y νyδy
⇐⇒ finding ψ : Y → R such that Tc#ψ ρ = ν
⇐⇒ maximizing the concave function Φ
Aurenhammer, Hoffman, Aronov ’98
Φ(ψ) := P
y
R
Vorψc (y)[c(x, y) + ψ(y)] d ρ(x) − P
y ψ(y)νy
Optimal transport as concave maximization
Theorem: Finding a c-O.T between ρ with density and ν = P
Y νyδy
⇐⇒ finding ψ : Y → R such that Tc#ψ ρ = ν
⇐⇒ maximizing the concave function Φ
Aurenhammer, Hoffman, Aronov ’98
Φ(ψ) := P
y
R
Vorψc (y)[c(x, y) + ψ(y)] d ρ(x) − P
y ψ(y)νy
I Byproduct of Kantorovich duality.
Optimal transport as concave maximization
Theorem: Finding a c-O.T between ρ with density and ν = P
Y νyδy
⇐⇒ finding ψ : Y → R such that Tc#ψ ρ = ν
⇐⇒ maximizing the concave function Φ
Aurenhammer, Hoffman, Aronov ’98
Φ(ψ) := P
y
R
Vorψc (y)[c(x, y) + ψ(y)] d ρ(x) − P
y ψ(y)νy
I Byproduct of Kantorovich duality.
I ∇Φ(ψ) = (ρ(Vorψc (y)) − νy)y∈Y . Hence,
∇Φ = 0 ⇐⇒ discrete Monge-Amp`ere equation:
∀y ∈ Y, ρ(Vorψc (y)) = νy.
Optimal transport as concave maximization
Theorem: Finding a c-O.T between ρ with density and ν = P
Y νyδy
⇐⇒ finding ψ : Y → R such that Tc#ψ ρ = ν
⇐⇒ maximizing the concave function Φ
Aurenhammer, Hoffman, Aronov ’98
Φ(ψ) := P
y
R
Vorψc (y)[c(x, y) + ψ(y)] d ρ(x) − P
y ψ(y)νy
I Byproduct of Kantorovich duality.
I ∇Φ(ψ) = (ρ(Vorψc (y)) − νy)y∈Y . Hence,
∇Φ = 0 ⇐⇒ discrete Monge-Amp`ere equation:
I Existing numerical methods: coordinate-wise increment with minimum step,
N3
∀y ∈ Y, ρ(Vorψc (y)) = νy.
Optimal transport as concave maximization
Theorem: Finding a c-O.T between ρ with density and ν = P
Y νyδy
⇐⇒ finding ψ : Y → R such that Tc#ψ ρ = ν
⇐⇒ maximizing the concave function Φ
Aurenhammer, Hoffman, Aronov ’98
Φ(ψ) := P
y
R
Vorψc (y)[c(x, y) + ψ(y)] d ρ(x) − P
y ψ(y)νy
I Byproduct of Kantorovich duality.
I ∇Φ(ψ) = (ρ(Vorψc (y)) − νy)y∈Y . Hence,
∇Φ = 0 ⇐⇒ discrete Monge-Amp`ere equation:
I Existing numerical methods: coordinate-wise increment with minimum step,
[Oliker–Prussner]
with complexity O(Nε3 log(N)), ε = precision.
Contribution: Efficient heuristic to optimize Φ −→ quasi-Newton and multiscale
∀y ∈ Y, ρ(Vorψc (y)) = νy.
Multiscale Approach to Semi-discrete OT, c = 1 2 k.k 2
Φν(ψ) := P
y
R
Vorψc (y)[c(x, y) + ψ(y)] d ρ(x) − P
y ψ(y)νy I Single scale
−→ Evaluation of Φ: Voronoi cells (CGAL) + Exact integration
−→ low-storage quasi-Newton method
[M., Comput. Graph. Forum / SGP 2011]
Multiscale Approach to Semi-discrete OT, c = 1 2 k.k 2
I Multiscale. (A) Simplification of target measure using Lloyd’s algorithm
ν0
Φν(ψ) := P
y
R
Vorψc (y)[c(x, y) + ψ(y)] d ρ(x) − P
y ψ(y)νy I Single scale
−→ Evaluation of Φ: Voronoi cells (CGAL) + Exact integration
−→ low-storage quasi-Newton method
[M., Comput. Graph. Forum / SGP 2011]
Multiscale Approach to Semi-discrete OT, c = 1 2 k.k 2
I Multiscale. (A) Simplification of target measure using Lloyd’s algorithm
ν0 ν1
π01 Φν(ψ) := P
y
R
Vorψc (y)[c(x, y) + ψ(y)] d ρ(x) − P
y ψ(y)νy I Single scale
−→ Evaluation of Φ: Voronoi cells (CGAL) + Exact integration
−→ low-storage quasi-Newton method
[M., Comput. Graph. Forum / SGP 2011]
Multiscale Approach to Semi-discrete OT, c = 1 2 k.k 2
I Multiscale. (A) Simplification of target measure using Lloyd’s algorithm
ν0 ν1 ν2
π01 π12
Φν(ψ) := P
y
R
Vorψc (y)[c(x, y) + ψ(y)] d ρ(x) − P
y ψ(y)νy I Single scale
−→ Evaluation of Φ: Voronoi cells (CGAL) + Exact integration
−→ low-storage quasi-Newton method
[M., Comput. Graph. Forum / SGP 2011]
Multiscale Approach to Semi-discrete OT, c = 1 2 k.k 2
I Multiscale. (A) Simplification of target measure using Lloyd’s algorithm
ν0 ν1 ν2
π01 π12
ψ2 = arg max Φν2 starting from ψ20 = 0
(B) Coarse-to-fine
Φν(ψ) := P
y
R
Vorψc (y)[c(x, y) + ψ(y)] d ρ(x) − P
y ψ(y)νy I Single scale
resolution:
−→ Evaluation of Φ: Voronoi cells (CGAL) + Exact integration
−→ low-storage quasi-Newton method
[M., Comput. Graph. Forum / SGP 2011]
Multiscale Approach to Semi-discrete OT, c = 1 2 k.k 2
I Multiscale. (A) Simplification of target measure using Lloyd’s algorithm
ν0 ν1 ν2
π01 π12
ψ2 = arg max Φν2 ψ1 = arg min Φν1
w. ψ10 = π12−1 ◦ ψ2 starting from ψ20 = 0
(B) Coarse-to-fine
Φν(ψ) := P
y
R
Vorψc (y)[c(x, y) + ψ(y)] d ρ(x) − P
y ψ(y)νy I Single scale
resolution:
−→ Evaluation of Φ: Voronoi cells (CGAL) + Exact integration
−→ low-storage quasi-Newton method
[M., Comput. Graph. Forum / SGP 2011]
Multiscale Approach to Semi-discrete OT, c = 1 2 k.k 2
I Multiscale. (A) Simplification of target measure using Lloyd’s algorithm
ν0 ν1 ν2
π01 π12
ψ2 = arg max Φν2 ψ1 = arg min Φν1
w. ψ10 = π12−1 ◦ ψ2 starting from ψ20 = 0 ψ0 = arg min Φν0
w. ψ00 = π01−1 ◦ ψ1
(B) Coarse-to-fine
Φν(ψ) := P
y
R
Vorψc (y)[c(x, y) + ψ(y)] d ρ(x) − P
y ψ(y)νy I Single scale
resolution:
−→ Evaluation of Φ: Voronoi cells (CGAL) + Exact integration
−→ low-storage quasi-Newton method
[M., Comput. Graph. Forum / SGP 2011]
Multiscale Approach to Semi-discrete OT, c = 1 2 k.k 2
I Multiscale. (A) Simplification of target measure using Lloyd’s algorithm
ν0 ν1 ν2
π01 π12
ψ2 = arg max Φν2 ψ1 = arg min Φν1
w. ψ10 = π12−1 ◦ ψ2 starting from ψ20 = 0 ψ0 = arg min Φν0
w. ψ00 = π01−1 ◦ ψ1
(B) Coarse-to-fine
Φν(ψ) := P
y
R
Vorψc (y)[c(x, y) + ψ(y)] d ρ(x) − P
y ψ(y)νy I Single scale
resolution:
−→ Evaluation of Φ: Voronoi cells (CGAL) + Exact integration
First numerical method able to handle large semi-discrete OT problems
−→ low-storage quasi-Newton method
[M., Comput. Graph. Forum / SGP 2011]
Some Pictures of Optimal Transport Plans, c = 1 2 k.k 2
Source: picture ”Cameraman”
Target: Lloyd sampling of picture ”Peppers” (k = 625)
McCann ’97 Displacement interpolation
[M., Comput. Graph. Forum / SGP 2011]
Some Pictures of Optimal Transport Plans, c = 1 2 k.k 2
The mass of Dirac at p is spread onto Vorψc (p)
Source: picture ”Cameraman”
Target: Lloyd sampling of picture ”Peppers” (k = 625)
McCann ’97 Displacement interpolation
ψ = (1 − t) · ψsol + t · 0 t = 14
[M., Comput. Graph. Forum / SGP 2011]
Some Pictures of Optimal Transport Plans, c = 1 2 k.k 2
Source: picture ”Cameraman”
Target: Lloyd sampling of picture ”Peppers” (k = 625)
McCann ’97 Displacement interpolation
[M., Comput. Graph. Forum / SGP 2011]
Some Pictures of Optimal Transport Plans, c = 1 2 k.k 2
The mass of Dirac at p is spread onto Vorψc (p)
Source: picture ”Cameraman”
Target: Lloyd sampling of picture ”Peppers” (k = 625)
McCann ’97 Displacement interpolation
ψ = (1 − t) · ψsol + t · 0 t = 34
[M., Comput. Graph. Forum / SGP 2011]
Some Pictures of Optimal Transport Plans, c = 1 2 k.k 2
Source: picture ”Cameraman”
Target: Lloyd sampling of picture ”Peppers” (k = 625)
McCann ’97 Displacement interpolation
[M., Comput. Graph. Forum / SGP 2011]
Some Pictures of Optimal Transport Plans, c = 1 2 k.k 2
k = 625
k = 15000
[M., Comput. Graph. Forum / SGP 2011]
Some Pictures of Optimal Transport Plans, c = 1 2 k.k 2
k = 625
k = 15000
[M., Comput. Graph. Forum / SGP 2011]
Some Pictures of Optimal Transport Plans, c = 1 2 k.k 2
k = 625
k = 15000
[M., Comput. Graph. Forum / SGP 2011]
3D version of the algorithm by B. L´evy (2014), with k up to 1M points.
Hessian of Kantorovich functional, c = 1 2 k.k 2
∇Φ(ψ) = (ρ(Vorψc (y)) − νy)y∈Y
Assume X ⊆ Rd is open, bounded and connected / ρ ∈ Pac(X) ∩ C0(Ω), ρ ≥ ε.
Hessian of Kantorovich functional, c = 1 2 k.k 2
Proposition: Φ is twice differentiable almost everywhere and
∂2Φ
∂z∂y (ψ) = R
Vorψc (y,z)
ρ(x) dx ky−zk
∂2Φ
∂y2 (ψ) = − P
z6=y
∂2Φ
∂y∂z (ψ)
∇Φ(ψ) = (ρ(Vorψc (y)) − νy)y∈Y
Assume X ⊆ Rd is open, bounded and connected / ρ ∈ Pac(X) ∩ C0(Ω), ρ ≥ ε.
Vorψc (y, z) := Vorψc (y) ∩ Vorψc (z)
Hessian of Kantorovich functional, c = 1 2 k.k 2
Proposition: Φ is twice differentiable almost everywhere and
∂2Φ
∂z∂y (ψ) = R
Vorψc (y,z)
ρ(x) dx ky−zk
∂2Φ
∂y2 (ψ) = − P
z6=y
∂2Φ
∂y∂z (ψ)
∇Φ(ψ) = (ρ(Vorψc (y)) − νy)y∈Y
Assume X ⊆ Rd is open, bounded and connected / ρ ∈ Pac(X) ∩ C0(Ω), ρ ≥ ε.
(y, z) ∈ G ⇐⇒ ∂z∂y∂2Φ (ψ) > 0
Vorψc (y, z) := Vorψc (y) ∩ Vorψc (z)
I D2Φ(ψ) is the Laplacian of a weighted graph G
Hessian of Kantorovich functional, c = 1 2 k.k 2
Proposition: Φ is twice differentiable almost everywhere and
∂2Φ
∂z∂y (ψ) = R
Vorψc (y,z)
ρ(x) dx ky−zk
∂2Φ
∂y2 (ψ) = − P
z6=y
∂2Φ
∂y∂z (ψ)
∇Φ(ψ) = (ρ(Vorψc (y)) − νy)y∈Y
Assume X ⊆ Rd is open, bounded and connected / ρ ∈ Pac(X) ∩ C0(Ω), ρ ≥ ε.
(y, z) ∈ G ⇐⇒ ∂z∂y∂2Φ (ψ) > 0 ⇐⇒ Vorψc (y, z) ∩ X 6= ∅.
Vorψc (y, z) := Vorψc (y) ∩ Vorψc (z)
i.e. G = 1-skeleton of restricted regular triangulation I D2Φ(ψ) is the Laplacian of a weighted graph G
Hessian of Kantorovich functional, c = 1 2 k.k 2
Proposition: Φ is twice differentiable almost everywhere and
∂2Φ
∂z∂y (ψ) = R
Vorψc (y,z)
ρ(x) dx ky−zk
∂2Φ
∂y2 (ψ) = − P
z6=y
∂2Φ
∂y∂z (ψ)
∇Φ(ψ) = (ρ(Vorψc (y)) − νy)y∈Y
Assume X ⊆ Rd is open, bounded and connected / ρ ∈ Pac(X) ∩ C0(Ω), ρ ≥ ε.
(y, z) ∈ G ⇐⇒ ∂z∂y∂2Φ (ψ) > 0 ⇐⇒ Vorψc (y, z) ∩ X 6= ∅.
Vorψc (y, z) := Vorψc (y) ∩ Vorψc (z)
i.e. G = 1-skeleton of restricted regular triangulation I D2Φ(ψ) is the Laplacian of a weighted graph G
I miny∈Y ρ(Vorψc (y) ∩ X) > 0 =⇒ G is connected
Hessian of Kantorovich functional, c = 1 2 k.k 2
Proposition: Φ is twice differentiable almost everywhere and
∂2Φ
∂z∂y (ψ) = R
Vorψc (y,z)
ρ(x) dx ky−zk
∂2Φ
∂y2 (ψ) = − P
z6=y
∂2Φ
∂y∂z (ψ)
∇Φ(ψ) = (ρ(Vorψc (y)) − νy)y∈Y
Assume X ⊆ Rd is open, bounded and connected / ρ ∈ Pac(X) ∩ C0(Ω), ρ ≥ ε.
(y, z) ∈ G ⇐⇒ ∂z∂y∂2Φ (ψ) > 0 ⇐⇒ Vorψc (y, z) ∩ X 6= ∅.
Vorψc (y, z) := Vorψc (y) ∩ Vorψc (z)
−→ local quadratic convergence of Newton’s method i.e. G = 1-skeleton of restricted regular triangulation I D2Φ(ψ) is the Laplacian of a weighted graph G
=⇒ KerD2Ψ(ψ) = {cst}
I miny∈Y ρ(Vorψc (y) ∩ X) > 0 =⇒ G is connected
Future work: Newton’s Method and Other Costs
I Quantitative lower bound on λ1(−D2Φ) using a discrete Cheeger inequality:
λ1(X) ≥ 14h(X)2 h(X) := minA min(HHdd−1(A),H(∂A)d(X\A)
where A ⊆ X and ∂A smooth
[ongoing work with Kitagawa and Thibert]
Future work: Newton’s Method and Other Costs
I Quantitative lower bound on λ1(−D2Φ) using a discrete Cheeger inequality:
λ1(X) ≥ 14h(X)2 h(X) := minA min(HHdd−1(A),H(∂A)d(X\A)
where A ⊆ X and ∂A smooth
I Possible improvement if one is able to control maxy diam(Vorψc (y))
[ongoing work with Kitagawa and Thibert]
' regularity result for discrete solutions.
Future work: Newton’s Method and Other Costs
I Quantitative lower bound on λ1(−D2Φ) using a discrete Cheeger inequality:
λ1(X) ≥ 14h(X)2 h(X) := minA min(HHdd−1(A),H(∂A)d(X\A)
where A ⊆ X and ∂A smooth
I Possible improvement if one is able to control maxy diam(Vorψc (y))
I Extension to Ma-Trudinger-Wang costs, using work from Kitagawa? [Kitagawa ’12]
[ongoing work with Kitagawa and Thibert]
' regularity result for discrete solutions.
Future work: Newton’s Method and Other Costs
Towards the global convergence of a damped Newton method for MTW costs ? I Quantitative lower bound on λ1(−D2Φ) using a discrete Cheeger inequality:
λ1(X) ≥ 14h(X)2 h(X) := minA min(HHdd−1(A),H(∂A)d(X\A)
where A ⊆ X and ∂A smooth
I Possible improvement if one is able to control maxy diam(Vorψc (y))
I Extension to Ma-Trudinger-Wang costs, using work from Kitagawa? [Kitagawa ’12]
[ongoing work with Kitagawa and Thibert]
' regularity result for discrete solutions.
3. Far-Field Reflector Problem
c(x, y ) = − log(1 − hx|y i) on S
2in collaboration with P. Machado and B. Thibert
Far-Field Reflector Problem
S∞2 Punctual light at origin o, density f on So2 Prescribed far-field: density g on S∞2
So2 o
o f
g
Far-Field Reflector Problem
S∞2 Punctual light at origin o, density f on So2 Prescribed far-field: density g on S∞2
So2
Goal: Find a surface R which sends (So2, f) to (S∞, g) under reflection by Snell-Descartes law.
o o
R
f
g
Far-Field Reflector Problem
S∞2 Punctual light at origin o, density f on So2 Prescribed far-field: density g on S∞2
So2
Goal: Find a surface R which sends (So2, f) to (S∞, g) under reflection by Snell-Descartes law.
o o
R
Generalized Monge-Amp`ere equation: g(TR(u)) det(DTR(u)) = f(u) f
g
TR maps the direction of an emmited ray in So2 to the direction of the reflected ray in S∞2 .
Far-Field Reflector Problem
S∞2 Punctual light at origin o, density f on So2 Prescribed far-field: density g on S∞2
So2
Goal: Find a surface R which sends (So2, f) to (S∞, g) under reflection by Snell-Descartes law.
o o
R
Generalized Monge-Amp`ere equation: g(TR(u)) det(DTR(u)) = f(u) f
g
TR maps the direction of an emmited ray in So2
I Fully non linear PDE over the sphere with BV2 boundary conditions
to the direction of the reflected ray in S∞2 .
Far-Field Reflector Problem: Semi-discrete case
S∞2 y1
y2
y3
Punctual light at origin o, µ ∈ Pac(So2) Prescribed far-field: ν = P
i νiδyi ∈ P(S∞2 )
So2 o
Far-Field Reflector Problem: Semi-discrete case
S∞2 y1
y2
y3
Punctual light at origin o, µ ∈ Pac(So2) Prescribed far-field: ν = P
i νiδyi ∈ P(S∞2 )
So2
Goal: Find a surface R which sends (So2, µ) to (S∞, ν) under reflection by Snell’s law.
o
R
Far-Field Reflector Problem: Semi-discrete case
y1
y2
y3
Punctual light at origin o, µ ∈ Pac(So2) Prescribed far-field: ν = P
i νiδyi ∈ P(S∞2 )
−→ Pi(κi) = solid paraboloid of revolution with
focal o, direction yi and focal distance κi Goal: Find a surface R which sends (So2, µ) to
(S∞, ν) under reflection by Snell’s law.
o P3
P2
µ
P1
So2
Far-Field Reflector Problem: Semi-discrete case
Punctual light at origin o, µ ∈ Pac(So2) Prescribed far-field: ν = P
i νiδyi ∈ P(S∞2 )
−→ Pi(κi) = solid paraboloid of revolution with
focal o, direction yi and focal distance κi
−→ PIκi = πS2
o ∩Nj=1Pj(κj)
∩ ∂Pi(κi)
Goal: Find a surface R which sends (So2, µ) to (S∞, ν) under reflection by Snell’s law.
PIκ3
o P3
P2
µ
= partition of So2 So2
Far-Field Reflector Problem: Semi-discrete case
Punctual light at origin o, µ ∈ Pac(So2) Prescribed far-field: ν = P
i νiδyi ∈ P(S∞2 )
−→ Pi(κi) = solid paraboloid of revolution with
focal o, direction yi and focal distance κi
−→ PIκi = πS2
o ∩Nj=1Pj(κj)
∩ ∂Pi(κi)
Discrete Monge-Amp`ere equation: Find ~κ such that ∀i, µ(PIκi ) = νi.
Goal: Find a surface R which sends (So2, µ) to (S∞, ν) under reflection by Snell’s law.
PIκ3
amount of light reflected in direction yi.
o P3
P2
µ
= partition of So2 So2
Caffarelli-Kochengin-Oliker ’99
(FF)
Semi-Discrete Far-Field Reflector Problem as OT
Lemma: With c(x, y) = − log(1 − hx|yi), and ψi := log(κi), PIκi = Vorψc (yi).
PIκ3
o
P3(κ3)
Wang ’04
Glimm-Oliker ’03 [Machado, M., Thibert, Symp. Comp. Geom 2014]
Semi-Discrete Far-Field Reflector Problem as OT
Lemma: With c(x, y) = − log(1 − hx|yi), and ψi := log(κi), PIκi = Vorψc (yi).
PIκ3
o
P3(κ3)
Wang ’04
Glimm-Oliker ’03 [Machado, M., Thibert, Symp. Comp. Geom 2014]
(FF) Find κ such that ∀i, µ(PIκi ) = νi. Corollary:
Semi-Discrete Far-Field Reflector Problem as OT
Lemma: With c(x, y) = − log(1 − hx|yi), and ψi := log(κi), PIκi = Vorψc (yi).
PIκ3
o
P3(κ3)
Wang ’04
Glimm-Oliker ’03 [Machado, M., Thibert, Symp. Comp. Geom 2014]
(FF) Find κ such that ∀i, µ(PIκi ) = νi.
⇐⇒
(OT) Find ψ such that µ(Vorψc (yi)) = νi Corollary:
Semi-Discrete Far-Field Reflector Problem as OT
Lemma: With c(x, y) = − log(1 − hx|yi), and ψi := log(κi), PIκi = Vorψc (yi).
PIκ3
o
P3(κ3)
Wang ’04
Glimm-Oliker ’03 [Machado, M., Thibert, Symp. Comp. Geom 2014]
I (FF) is a semi-discrete OT problem with c(x, y) = − log(1 − hx|yi)
(FF) Find κ such that ∀i, µ(PIκi ) = νi.
⇐⇒
(OT) Find ψ such that µ(Vorψc (yi)) = νi Corollary:
Semi-Discrete Far-Field Reflector Problem as OT
Lemma: With c(x, y) = − log(1 − hx|yi), and ψi := log(κi), PIκi = Vorψc (yi).
PIκ3
o
P3(κ3)
Wang ’04
Glimm-Oliker ’03 [Machado, M., Thibert, Symp. Comp. Geom 2014]
I (FF) is a semi-discrete OT problem with c(x, y) = − log(1 − hx|yi)
(FF) Find κ such that ∀i, µ(PIκi ) = νi.
⇐⇒
(OT) Find ψ such that µ(Vorψc (yi)) = νi Corollary:
Computation of the generalized Voronoi cells
Lemma: With ψi = log(κi), pi := −2κyi
i and ωi := −k2κyi
i k2 − κ1
i , PIκi = Vorψc (yi) = VorωP,k.k2(pi) ∩ S2
[Machado, M., Thibert, Symp. Comp. Geom 2014]
Computation of the generalized Voronoi cells
Lemma: With ψi = log(κi), pi := −2κyi
i and ωi := −k2κyi
i k2 − κ1
i , PIκi = Vorψc (yi) = VorωP,k.k2(pi) ∩ S2
I Computation combines Regular triangulation 3 from
[Machado, M., Thibert, Symp. Comp. Geom 2014]
CGAL with custom filtered predicates and data structures.
Computation of the generalized Voronoi cells
Lemma: With ψi = log(κi), pi := −2κyi
i and ωi := −k2κyi
i k2 − κ1
i , PIκi = Vorψc (yi) = VorωP,k.k2(pi) ∩ S2
I Computation combines Regular triangulation 3 from
[Machado, M., Thibert, Symp. Comp. Geom 2014]
I Applies to other geometric constructions involving confocal CGAL with custom filtered predicates and data structures.
quadrics of revolution appearing in geometric optics.