Matthew Wright Institute for Mathematics and its Applications University of Minnesota Applied Topology in Bฤdlewo July 24, 2013 How can we assign a notion of size to functions? Lebesgue integral Anything else? Euler Characteristic Let ๐ be a finite simplicial complex containing ๐ด๐ open simplices of dimension ๐. ๐ด0 = number of vertices of ๐ด ๐ด1 = number of edges of ๐ด ๐ด2 = number of faces of ๐ด etc. Then the Euler Characteristic of ๐ด is: v combinatorial −1 ๐ ๐ด๐ ๐ ๐ด = ๐ Key Property For sets ๐ด and ๐ต, ๐ ๐ด∪๐ต =๐ ๐ด +๐ ๐ต −๐ ๐ด∩๐ต . This property is called additivity, or the inclusion-exclusion principle. ๐ด ๐ด∩๐ต ๐ต Euler Integral Let ๐ด be a “tame” set in โ๐ , and let ๐๐ด be the function with value 1 on set ๐ด and 0 otherwise. The Euler Integral of ๐๐ด is: โ๐ ๐๐ด ๐๐ = ๐(๐ด) For a “tame” function ๐: โ๐ → โค, with finite range, โ๐ ๐ ๐๐ = ๐ ๐{๐ = ๐} . ๐ set on which ๐ = ๐ Example Consider ๐: โ → โค: 3 ๐(๐ฅ) 2 1 โ๐ ๐ ๐๐ = ๐ ๐{๐ = ๐} ๐ =1⋅0 ←๐=1 + 2 ⋅ (−1) ←๐=2 +3⋅2 ←๐=3 =4 ๐ฅ Euler integral of ๐ Continuous Functions How can we extend the Euler integral to a continuous function ๐: โ → โ? Idea: Approximate ๐ by step functions. 3 ๐ ๐ Make the step size smaller. Consider the limit of the 2๐ Euler integrals of the 2 approximations as the 1 step size goes to zero: 1 lim ๐๐ ๐๐ ๐ฅ ๐→∞ ๐ Does it matter if we use lower or upper approximations? 1 โ 2 Continuous Functions To extend the Euler integral to a function ๐: โ๐ → โ, define two integrals: 1 ๐ ๐๐ = lim ๐๐ ๐๐ Lower integral: ๐→∞ ๐ Upper integral: 1 ๐ ๐๐ = lim ๐→∞ ๐ ๐๐ ๐๐ These limits exist, but are not equal in general. Application Local Data Global Data Euler Integration is useful in sensor networks: • Networks of cell phones or computers • Traffic sensor networks • Surveillance and radar networks How can we assign a notion of size to functions? Lebesgue integral Euler integral Anything else? Intrinsic Volumes The intrinsic volumes are the ๐ + 1 Euclidean-invariant valuations on subsets of โ๐ , denoted ๐0 , … , ๐๐ . ๐0 : Euler characteristic 1 ๐1 : “length” ๐๐−1 : ½(surface area) 0 ๐๐ : (Lebesgue) volume ๐ = ๐๐คโ Example Let ๐พ be an ๐-dimensional closed box with side lengths ๐ฅ1 , ๐ฅ2 , … , ๐ฅ๐ . The ๐ th intrinsic volume of ๐พ is ๐๐ (๐ฅ1 , ๐ฅ2 , … , ๐ฅ๐ ), the elementary symmetric polynomial of degree ๐ on ๐ variables. ๐0 ๐พ = ๐0 ๐ฅ1 , … , ๐ฅ๐ = 1 ๐ฅ3 ๐1 ๐พ = ๐1 ๐ฅ1 , … , ๐ฅ๐ = ๐ฅ1 + ๐ฅ2 + โฏ + ๐ฅ๐ ๐2 ๐พ = ๐2 (๐ฅ1 , … , ๐ฅ๐ ) = ๐ฅ1 ๐ฅ2 + ๐ฅ1 ๐ฅ3 + โฏ + ๐ฅ๐−1 ๐ฅ๐ โฎ ๐๐ ๐พ = ๐๐ ๐ฅ1 , … , ๐ฅ๐ = ๐ฅ1 ๐ฅ2 โฏ ๐ฅ๐ ๐ฅ1 ๐ฅ2 Intrinsic Volume Definition For a “tame” set ๐พ ⊂ โ, the ๐th intrinsic volume can be defined: Hadwiger’s Formula ๐๐ ๐พ = ๐ ๐พ ∩ ๐ ๐๐(๐) ๐ด๐,๐−๐ ๐ด๐,๐−๐ is the affine Grassmanian of (๐ − ๐)– dimensional planes in โ๐ , and ๐ is Harr measure on ๐ด๐,๐−๐ with appropriate normalization. Tube Formula tube(๐พ, ๐) ๐พ ๐ The volume of a tube around ๐พ is a polynomial in ๐, whose coefficients involve intrinsic volumes of ๐พ. Steiner Formula: For compact convex ๐พ ⊂ โ๐ and ๐ > 0, ๐ ๐๐−๐ ๐๐ (๐พ)๐ ๐−๐ ๐๐ (tube ๐พ, ๐ ) = ๐=0 volume of unit (๐ − ๐)-ball intrinsic volume Hadwiger Integral Let ๐ โถ โ๐ → โค have finite range. Integration of ๐ with respect to ๐๐ is straightforward: โ๐ ๐ ๐๐๐ = ๐ ๐๐ {๐ = ๐} set on which ๐=๐ ๐ Integration of ๐ โถ โ๐ → โ is more complicated: Lower integral: โ๐ Upper integral: โ๐ ๐ ๐๐๐ 1 = lim ๐→∞ ๐ ๐ ๐๐๐ 1 = lim ๐→∞ ๐ โ๐ โ๐ ๐๐ ๐๐๐ ๐๐ ๐๐๐ Hadwiger Integral Let ๐ ⊆ โ๐ be compact and ๐ โถ ๐ → โ+ bounded. ∞ ๐ ๐๐๐ = X ๐๐ ๐ ≥ ๐ ๐๐ = s=0 ๐ ๐๐ ๐๐พ ๐ด๐,๐−๐ ๐ ∩ ๐ slices level sets ๐ ๐ Example Let ๐ ๐ฅ, ๐ฆ = 4 − ๐ฅ 2 − ๐ฆ 2 on ๐ = ๐ฅ, ๐ฆ | ๐ฅ 2 − ๐ฆ 2 ≤ 4 . Excursion set ๐ ≥ ๐ is a circle of radius 4 − ๐ . ๐ Hadwiger Integrals: 4 ๐ ๐ ๐ ๐๐0 = 1 ๐๐ = 4 0 4 ๐ ๐ ๐ ๐๐1 = 0 16๐ ๐ 4 − ๐ ๐๐ = 3 4 ๐ ๐ ๐๐2 = ๐(4 − ๐ ) ๐๐ = 8๐ 0 Valuations on Functions A valuation on functions is an additive map ๐ฃ โถ {“tame” functions on โ๐} → โ. For a valuation on functions, additivity means ๐ฃ(๐ ∨ ๐) + ๐ฃ(๐ ∧ ๐) = ๐ฃ(๐ ) + ๐ฃ(๐), pointwise max pointwise min or equivalently, ๐ฃ(๐ ) = ๐ฃ(๐ ⋅ ๐๐ด ) + ๐ฃ(๐ ⋅ ๐๐ด๐ ) for any subset ๐ด and its complement ๐ด๐ . Valuations on Functions A valuation on functions is an additive map ๐ฃ โถ {“tame” functions on โ๐} → โ. Valuation ๐ฃ is: • Euclidean-invariant if ๐ฃ(๐ ) = ๐ฃ(๐(๐)) for any Euclidean motion ๐ of โ๐ . • continuous if a “small” change in ๐ corresponds to a “small” change in ๐ฃ(๐) (a precise definition of continuity requires a discussion of the flat topology on functions). Hadwiger’s Theorem for Functions (Baryshnikov, Ghrist, Wright) Any Euclidean-invariant, continuous valuation ๐ฃ on “tame” functions can be written ๐ ๐ฃ ๐ = ๐ โ ๐=0 ๐๐ ๐ ๐๐๐ for some increasing functions ๐๐ : โ → โ. That is, any valuation on functions can be written as a sum of Hadwiger integrals. How can we assign a notion of size to functions? Lebesgue integral Euler integral Hadwiger Integral Any valuation on functions can be written in terms of Hadwiger integrals. Surveillance ๐ 3 2 0 1 2 0 0 1 2 1 3 2 3 2 1 1 0 2 1 Suppose function ๐ counts the number of objects at each point in a domain. Hadwiger integrals provide data about the set of objects: ๐ ๐๐0 gives a count ๐ ๐๐1 gives a “length” ๐ ๐๐2 gives an “area” etc. Cell Dynamics As the cell structure changes by a certain process that minimizes energy, cell volumes change according to: ๐๐๐ 1 ๐ถ = −2๐๐ ๐๐−2 ๐ถ๐ − ๐๐−2 (๐ถ๐−2 ) ๐๐ก 6 ๐-dimensional structure (๐ − 2)-dimensional structure Image Processing Intrinsic volumes are of utility in image processing. A greyscale image can be viewed as a real-valued function on a planar domain. With such a perspective, Hadwiger integrals may be useful to return information about an image. Applications may also include color or hyperspectral images, or images on higher-dimensional domains. Percolation Question: Can liquid flow through a porous material from top to bottom? โ3 Functional approach: Define a permeability function in a solid material. Hadwiger integrals may be useful in such a functional approach to percolation theory. Surveillance Let ๐: ๐ → โค count objects locally in a domain ๐ ⊆ โ2 . ๐ 3 2 0 1 2 0 0 1 1 3 ? 2 ? 3 2 1 1 0 ?2 2 1 Then the Euler integral gives the global count: ๐ ๐ ๐๐0 = 5 What if part of ๐ is not observable? Idea: Model the function with a random field. Estimate the global count via the expected Euler integral. Random Field Intuitively: A random field is a function whose value at any point in its domain is a random variable. Formally: Let Ω, โฑ, โ be a probability space and ๐ a topological space. A measurable mapping ๐: Ω → โ๐ (the space of all real-valued functions on ๐) is called a realvalued random field. Note: ๐(๐) is a function, (๐ ๐ )(๐ก) is its value at ๐ก. Shorthand: Let ๐๐ก = (๐ ๐ )(๐ก). Expected Hadwiger Integral Theorem: Let ๐ โถ ๐ → โ๐ be a ๐-dimensional Gaussian field satisfying the conditions of the Gaussian Kinematic Formula. Let ๐น โถ โ๐ → โ be a piecewise ๐ถ 2 function. Let ๐ = ๐น โ ๐, so ๐ โถ ๐ → โ is a Gaussian-related field. Then the expected lower Hadwiger integral of ๐ is: ๐ผ ๐ ๐ ๐๐๐ dim ๐ −๐ ๐=1 = ๐๐ ๐ ๐ผ ๐ + ๐+๐ 2๐ ๐ −๐/2 ๐ ๐+๐ ๐พ ๐ โ โณ๐ {๐น ≥ ๐ข} ๐๐ข and similarly for the expected upper Hadwiger integral. Computational Difficulties Computing expected Hadwiger integrals of random fields is difficult in general. ๐ผ ๐ ๐ ๐๐๐ dim ๐ −๐ ๐=1 = ๐๐ ๐ ๐ผ ๐ + ๐+๐ 2๐ ๐ −๐/2 ๐๐+๐ ๐ โ ๐พ โณ๐ {๐น ≥ ๐ข} ๐๐ข intrinsic volumes: tricky, but possible to compute Gaussian Minkowski functionals: very difficult to compute, except in special cases Challenge: Non-Linearity Consider the following Euler integrals: ๐ฆ=1 1 1 ๐ฆ=๐ฅ ๐ฅ ๐ฅ ๐๐ = 1 [0, 1] 1 ๐ฆ =1−๐ฅ ๐ฅ (1 − ๐ฅ) ๐๐ = 1 [0, 1] ๐ฅ 1 ๐๐ = 1 [0, 1] Upper and lower Hadwiger integrals are not linear in general. Challenge: Continuity A change in a function ๐ on a small set (in the Lebesgue) sense can result in a large change in the Hadwiger integrals of ๐. 2 1 2 ๐ 1 ๐ ๐ฅ ๐ ๐๐ = 1 ๐ฅ Similar examples exist for higherdimensional Hadwiger integrals. ๐ ๐๐ = 2 Working with Hadwiger integrals requires different intuition than working with Lebesgue integrals. Challenge: Approximations How can we approximate the Hadwiger integrals of a function sampled at discrete points? ๐: 0,1 2 →โ triangulated approximations of ๐ Hadwiger integrals of interpolations of ๐ might diverge, even when the approximations converge pointwise to ๐. Summary • The intrinsic volumes provide notions of size for sets, generalizing both Euler characteristic and Lebesgue measure. • Analogously, the Hadwiger integrals provide notions of size for real-valued functions. • Hadwiger integrals are useful in applications such as surveillance, sensor networks, cell dynamics, and image processing. • Hadwiger integrals bring theoretical and computational challenges, and provide many open questions for future study. References • Yuliy Baryshnikov and Robert Ghrist. “Target Enumeration via Euler Characteristic Integration.” SIAM J. Appl. Math. 70(3), 2009, 825–844. • Yuliy Baryshnikov and Robert Ghrist. “Definable Euler integration.” Proc. Nat. Acad. Sci. 107(21), 2010, 9525-9530. • Yuliy Baryshnikov, Robert Ghrist, and Matthew Wright. “Hadwiger’s Theorem for Definable Functions.” Advances in Mathematics. Vol. 245 (2013) p. 573-586. • Omer Bobrowski and Matthew Strom Borman. “Euler Integration of Gaussian Random Fields and Persistent Homology.” Journal of Topology and Analysis, 4(1), 2012. • S. H. Shanuel. “What is the Length of a Potato?” Lecture Notes in Mathematics. Springer, 1986, 118 – 126. • Matthew Wright. “Hadwiger Integration of Definable Functions.” Publicly accessible Penn Dissertations. Paper 391. http://repository.upenn.edu/edissertations/391.