energy optimization with orthogonal potentials on the
play

Energy Optimization with Orthogonal Potentials on the Sphere Ryan - PowerPoint PPT Presentation

Energy Optimization with Orthogonal Potentials on the Sphere Ryan W. Matzke University of Minnesota - Twin Cities November 28, 2018 In collaboration with Dmitriy Bilyk, Alexey Glazyrin, Josiah Park, and Alex Vlasiuk Ryan W. Matzke Energy


  1. Energy Optimization with Orthogonal Potentials on the Sphere Ryan W. Matzke University of Minnesota - Twin Cities November 28, 2018 In collaboration with Dmitriy Bilyk, Alexey Glazyrin, Josiah Park, and Alex Vlasiuk Ryan W. Matzke Energy Optimization with Orthogonal Potentials on the Sphere

  2. Energy Given a potential function F : [ − 1 , 1 ] → R Ryan W. Matzke Energy Optimization with Orthogonal Potentials on the Sphere

  3. Energy Given a potential function F : [ − 1 , 1 ] → R The (discrete) energy of Z ⊂ S d , | Z | = N , with respect to F , is E F ( Z ) = 1 � F ( � x , y � ) . N 2 x , y ∈ Z Ryan W. Matzke Energy Optimization with Orthogonal Potentials on the Sphere

  4. Energy Given a potential function F : [ − 1 , 1 ] → R The (discrete) energy of Z ⊂ S d , | Z | = N , with respect to F , is E F ( Z ) = 1 � F ( � x , y � ) . N 2 x , y ∈ Z The (continuous) energy of a measure µ ∈ B ( S d ) , with respect to F , is � � I F ( µ ) = S d F ( � x , y � ) d µ ( x ) d µ ( y ) . S d Ryan W. Matzke Energy Optimization with Orthogonal Potentials on the Sphere

  5. Energy Given a potential function F : [ − 1 , 1 ] → R The (discrete) energy of Z ⊂ S d , | Z | = N , with respect to F , is E F ( Z ) = 1 � F ( � x , y � ) . N 2 x , y ∈ Z The (continuous) energy of a measure µ ∈ B ( S d ) , with respect to F , is � � I F ( µ ) = S d F ( � x , y � ) d µ ( x ) d µ ( y ) . S d 1 If µ = � x ∈ Z δ x , then | Z | I F ( µ ) = 1 � F ( � x , y � ) = E F ( Z ) . N 2 x , y ∈ Z Ryan W. Matzke Energy Optimization with Orthogonal Potentials on the Sphere

  6. Appearances of Monotonic Potentials Electrostatics (Riesz energy): F ( � x , y � ) = || x − y || − s . Ryan W. Matzke Energy Optimization with Orthogonal Potentials on the Sphere

  7. Appearances of Monotonic Potentials Electrostatics (Riesz energy): F ( � x , y � ) = || x − y || − s . Stolarsky Invariance Principle: F ( � x , y � ) = || x − y || . Ryan W. Matzke Energy Optimization with Orthogonal Potentials on the Sphere

  8. Appearances of Monotonic Potentials Electrostatics (Riesz energy): F ( � x , y � ) = || x − y || − s . Stolarsky Invariance Principle: F ( � x , y � ) = || x − y || . Packing Problem: � ∞ || x − y || < δ F ( � x , y � ) = || x − y || ≥ δ . 0 Ryan W. Matzke Energy Optimization with Orthogonal Potentials on the Sphere

  9. Orthogonal Potentials Frame Potential: F ( � x , y � ) = |� x , y �| 2 . Ryan W. Matzke Energy Optimization with Orthogonal Potentials on the Sphere

  10. Orthogonal Potentials Frame Potential: F ( � x , y � ) = |� x , y �| 2 . Fejes Tóth Conjecture (sum of acute angles): F ( � x , y � ) = 1 π arccos( |� x , y �| ) . Ryan W. Matzke Energy Optimization with Orthogonal Potentials on the Sphere

  11. Orthogonal Potentials Frame Potential: F ( � x , y � ) = |� x , y �| 2 . Fejes Tóth Conjecture (sum of acute angles): F ( � x , y � ) = 1 π arccos( |� x , y �| ) . p -Frame Potential: F ( � x , y � ) = |� x , y �| p . Ryan W. Matzke Energy Optimization with Orthogonal Potentials on the Sphere

  12. Orthogonal Potentials Frame Potential: F ( � x , y � ) = |� x , y �| 2 . Fejes Tóth Conjecture (sum of acute angles): F ( � x , y � ) = 1 π arccos( |� x , y �| ) . p -Frame Potential: F ( � x , y � ) = |� x , y �| p . These orthogonal potentials are monotonic in real projective space. Ryan W. Matzke Energy Optimization with Orthogonal Potentials on the Sphere

  13. Positive Definiteness Definition A function f : [ − 1 , 1 ] → R is positive definite on S d if, for all n ∈ N , x 1 , ..., x n ∈ S d , c 1 , ..., c n ∈ R , n n � � c i c j f ( � x i , x j � ) ≥ 0 . i = 1 j = 1 Ryan W. Matzke Energy Optimization with Orthogonal Potentials on the Sphere

  14. Positive Definiteness Definition A function f : [ − 1 , 1 ] → R is positive definite on S d if, for all n ∈ N , x 1 , ..., x n ∈ S d , c 1 , ..., c n ∈ R , n n � � c i c j f ( � x i , x j � ) ≥ 0 . i = 1 j = 1 A function f is positive definite iff for all n ∈ N � 1 f ( n ; d − 1 d − 1 d − 2 ˆ ( t )( 1 − t 2 ) ) = a d , n f ( t ) C 2 dt ≥ 0 . 2 n 2 − 1 Ryan W. Matzke Energy Optimization with Orthogonal Potentials on the Sphere

  15. Positive Definiteness Definition A function f : [ − 1 , 1 ] → R is positive definite on S d if, for all n ∈ N , x 1 , ..., x n ∈ S d , c 1 , ..., c n ∈ R , n n � � c i c j f ( � x i , x j � ) ≥ 0 . i = 1 j = 1 A function f is positive definite iff for all n ∈ N � 1 f ( n ; d − 1 d − 1 d − 2 ˆ ( t )( 1 − t 2 ) ) = a d , n f ( t ) C 2 dt ≥ 0 . 2 n 2 − 1 Theorem σ is a minimizer of I F ( µ ) if and only if F is positive definite. Ryan W. Matzke Energy Optimization with Orthogonal Potentials on the Sphere

  16. Negative Definiteness Definition A function f : [ − 1 , 1 ] → R is negative definite on S d if, for all n ∈ N , x 1 , ..., x n ∈ S d , c 1 , ..., c n ∈ R , n n � � c i c j f ( � x i , x j � ) ≤ 0 . i = 1 j = 1 A function f is negative definite iff for all n ∈ N � 1 f ( n ; d − 1 d − 1 d − 2 ˆ ( t )( 1 − t 2 ) 2 dt ≤ 0 . ) = a d , n f ( t ) C 2 n 2 − 1 Theorem σ is a maximizer of I F ( µ ) if and only if F is negative definite. Ryan W. Matzke Energy Optimization with Orthogonal Potentials on the Sphere

  17. Frame Potential Definition We call a finite set of unit vectors { x 1 , ..., x N } ⊂ S d a finite unit norm tight frame (FUNTF) if there exists some constant A > 0 such that for all y ∈ R d + 1 N |� y , x j �| 2 = A || y || 2 . � j = 1 Definition The frame potential of { x i } N i = 1 is N i = 1 ) = 1 FP ( { x i } N � |� x i , x j �| 2 , N 2 i , j = 1 Ryan W. Matzke Energy Optimization with Orthogonal Potentials on the Sphere

  18. Frame Potential Definition The frame potential of { x i } N i = 1 is N i = 1 ) = 1 � FP ( { x i } N |� x i , x j �| 2 . N 2 i , j = 1 The frame potential of µ ∈ B ( S d ) is � � S d |� x , y �| 2 d µ ( x ) d µ ( y ) . FP ( µ ) = S d Ryan W. Matzke Energy Optimization with Orthogonal Potentials on the Sphere

  19. Frame Potential Definition The frame potential of { x i } N i = 1 is N i = 1 ) = 1 � FP ( { x i } N |� x i , x j �| 2 . N 2 i , j = 1 The frame potential of µ ∈ B ( S d ) is � � S d |� x , y �| 2 d µ ( x ) d µ ( y ) . FP ( µ ) = S d Theorem (Benedetto, Fickus (2003)) 1 If N ≥ d + 1 , the minimum value of the frame potential is d + 1 , and the (local/global) minimizers are precisely the FUNTF’s in R d + 1 . Ryan W. Matzke Energy Optimization with Orthogonal Potentials on the Sphere

  20. Fejes Tóth Conjecture Conjecture (Fejes Tóth (1959)) Let d ≥ 1 , N = m ( d + 1 ) + k with m ∈ N 0 and 0 ≤ k ≤ d, and F ( � x , y � ) = 1 π arccos( |� x , y �| ) . Then E F ( Z ) is maximized by the point set Z = { z 1 , . . . , z N } ⊂ S d with z p ( d + 1 )+ i = e i . In this case, the energy is k ( k − 1 )( m + 1 ) 2 + 2 km ( d + 1 − k )( m + 1 ) + ( d − k )( d + 1 − k ) m 2 . 2 N 2 In particular, if N = m ( d + 1 ) , the sum is maximized by m copies of the orthonormal basis: E F ( Z ) = 1 d max 2 · d + 1 . Z ⊂ S d # Z = N Ryan W. Matzke Energy Optimization with Orthogonal Potentials on the Sphere

  21. Bounding the Energy G ( t ) = 1 2 − 69 50 π t 2 ≥ 1 π arccos( | t | ) = F ( t ) . Ryan W. Matzke Energy Optimization with Orthogonal Potentials on the Sphere

  22. Bounding the Energy G ( t ) = 1 2 − 69 50 π t 2 ≥ 1 π arccos( | t | ) = F ( t ) . Figure: The graph of the function G ( t ) − F ( t ) for 0 ≤ t ≤ 1. Ryan W. Matzke Energy Optimization with Orthogonal Potentials on the Sphere

  23. Bounding the Energy G ( t ) = 1 2 − 69 50 π t 2 ≥ 1 π arccos( | t | ) = F ( t ) . Figure: The graph of the function G ( t ) − F ( t ) for 0 ≤ t ≤ 1. From results of Benedetto and Fickus on frame potential, we have d µ ∈ B ( S d ) I G ( µ ) = 1 69 2 ( d + 1 ) ≤ µ ∈ B ( S d ) I F ( µ ) ≤ max max 2 − 50 π ( d + 1 ) . Ryan W. Matzke Energy Optimization with Orthogonal Potentials on the Sphere

  24. ✶ On the Circle Proofs: Geometric (Fodor, Vígh, Zarnocz), Fourier expansion, Chebyshev expansion. Ryan W. Matzke Energy Optimization with Orthogonal Potentials on the Sphere

  25. On the Circle Proofs: Geometric (Fodor, Vígh, Zarnocz), Fourier expansion, Chebyshev expansion. For x ∈ S 1 , define the antipodal quadrants in the direction of x as √ 2 Q ( x ) = { y : |� x , y �| > 2 } . We have a Quadrant Stolarsky Principle: Proposition (Bilyk, Matzke (2018)) For an N-point set Z ⊂ S 1 , 2 � N � � � 2 = 1 � � � � � � D L 2 , quad ( Z ) ✶ Q ( x ) ( z i ) − σ Q ( x ) d σ ( x ) � � � N � S 1 � � i = 1 N = 1 4 − 1 π · 1 � arccos |� z i , z j �| . N 2 i , j = 1 Ryan W. Matzke Energy Optimization with Orthogonal Potentials on the Sphere

  26. p-Frame Potential Definition For p ∈ ( 0 , ∞ ) , µ ∈ B ( S d ) , and Z ⊆ S d , we define the p-frame potential of µ as � � S d |� x , y �| p d µ ( x ) d µ ( y ) FP ( µ, p ) = S d and the p-frame potential of Z as 1 � |� x , y �| p . FP ( Z , p ) = | Z | 2 x , y ∈ Z Ryan W. Matzke Energy Optimization with Orthogonal Potentials on the Sphere

Recommend


More recommend