robustness in geometric computation
play

Robustness in Geometric Computation Vikram Sharma 1 1 Institute of - PowerPoint PPT Presentation

Jan 12, 2024 •274 likes •1.67k views

Nonrobustness Real Root Isolation Robustness in Geometric Computation Vikram Sharma 1 1 Institute of Mathematical Sciences, Chennai, India. Mysore Park Workshop, August, 2013 Nonrobustness Real Root Isolation Outline Nonrobustness 1 In

  1. Nonrobustness Real Root Isolation p 7 p 2 Algebraic form of orient p 3 Sign of (( q x − p x )( r y − p y ) − ( q y − p y )( r x − p x )) . ” is + 1, “R” is − 1, “S” is 0. “L p 8 In practice, fl_orient ( p , q , r ) . p 6 p 1 Is p 4 in ∆( p 1 , p 2 , p 3 ) ? 1 p 5 Is fl_orient ( p 1 , p 2 , p 4 ) = L ? Yes. p 4 2 Is fl_orient ( p 2 , p 3 , p 4 ) = L ? Yes. 3 p 1 = (7 . 3000000000000194 , 7 . 3000000000000167) p 2 = (24 . 000000000000068 , 24 . 000000000000071) Is fl_orient ( p 3 , p 1 , p 4 ) = L ? 4 p 3 = (24 . 00000000000005 , 24 . 000000000000053) p 4 = (0 . 50000000000001621 , 0 . 50000000000001243) Yes. 5 p 5 = (8 , 4) p 6 = (4 , 9) p 7 = (15 , 27) p 8 = (19 , 11) .

  2. Nonrobustness Real Root Isolation p 7 p 2 Algebraic form of orient p 3 Sign of (( q x − p x )( r y − p y ) − ( q y − p y )( r x − p x )) . ” is + 1, “R” is − 1, “S” is 0. “L p 8 In practice, fl_orient ( p , q , r ) . p 6 p 1 Is p 4 in ∆( p 1 , p 2 , p 3 ) ? 1 p 5 Is fl_orient ( p 1 , p 2 , p 4 ) = L ? Yes. p 4 2 Is fl_orient ( p 2 , p 3 , p 4 ) = L ? Yes. 3 p 1 = (7 . 3000000000000194 , 7 . 3000000000000167) p 2 = (24 . 000000000000068 , 24 . 000000000000071) Is fl_orient ( p 3 , p 1 , p 4 ) = L ? 4 p 3 = (24 . 00000000000005 , 24 . 000000000000053) p 4 = (0 . 50000000000001621 , 0 . 50000000000001243) Yes. 5 p 5 = (8 , 4) p 6 = (4 , 9) p 7 = (15 , 27) p 8 = (19 , 11) .

  3. Nonrobustness Real Root Isolation p 7 p 2 Algebraic form of orient p 3 Sign of (( q x − p x )( r y − p y ) − ( q y − p y )( r x − p x )) . ” is + 1, “R” is − 1, “S” is 0. “L p 8 In practice, fl_orient ( p , q , r ) . p 6 p 1 Is p 4 in ∆( p 1 , p 2 , p 3 ) ? 1 p 5 Is fl_orient ( p 1 , p 2 , p 4 ) = L ? Yes. p 4 2 Is fl_orient ( p 2 , p 3 , p 4 ) = L ? Yes. 3 p 1 = (7 . 3000000000000194 , 7 . 3000000000000167) p 2 = (24 . 000000000000068 , 24 . 000000000000071) Is fl_orient ( p 3 , p 1 , p 4 ) = L ? 4 p 3 = (24 . 00000000000005 , 24 . 000000000000053) p 4 = (0 . 50000000000001621 , 0 . 50000000000001243) Yes. 5 p 5 = (8 , 4) p 6 = (4 , 9) p 7 = (15 , 27) p 8 = (19 , 11) . Is p 5 in ∆( p 1 , p 2 , p 3 ) ? No. 6

  4. Nonrobustness Real Root Isolation p 7 p 2 Algebraic form of orient p 3 Sign of (( q x − p x )( r y − p y ) − ( q y − p y )( r x − p x )) . ” is + 1, “R” is − 1, “S” is 0. “L p 8 In practice, fl_orient ( p , q , r ) . p 6 p 1 Is p 4 in ∆( p 1 , p 2 , p 3 ) ? 1 p 5 Is fl_orient ( p 1 , p 2 , p 4 ) = L ? Yes. p 4 2 Is fl_orient ( p 2 , p 3 , p 4 ) = L ? Yes. 3 p 1 = (7 . 3000000000000194 , 7 . 3000000000000167) p 2 = (24 . 000000000000068 , 24 . 000000000000071) Is fl_orient ( p 3 , p 1 , p 4 ) = L ? 4 p 3 = (24 . 00000000000005 , 24 . 000000000000053) p 4 = (0 . 50000000000001621 , 0 . 50000000000001243) Yes. 5 p 5 = (8 , 4) p 6 = (4 , 9) p 7 = (15 , 27) p 8 = (19 , 11) . Is p 5 in ∆( p 1 , p 2 , p 3 ) ? No. 6 Update, and continue... 7

  5. Nonrobustness Real Root Isolation p 7 p 2 Algebraic form of orient p 3 Sign of (( q x − p x )( r y − p y ) − ( q y − p y )( r x − p x )) . ” is + 1, “R” is − 1, “S” is 0. “L p 8 In practice, fl_orient ( p , q , r ) . p 6 p 1 Is p 4 in ∆( p 1 , p 2 , p 3 ) ? 1 p 5 Is fl_orient ( p 1 , p 2 , p 4 ) = L ? Yes. p 4 2 Is fl_orient ( p 2 , p 3 , p 4 ) = L ? Yes. 3 p 1 = (7 . 3000000000000194 , 7 . 3000000000000167) p 2 = (24 . 000000000000068 , 24 . 000000000000071) Is fl_orient ( p 3 , p 1 , p 4 ) = L ? 4 p 3 = (24 . 00000000000005 , 24 . 000000000000053) p 4 = (0 . 50000000000001621 , 0 . 50000000000001243) Yes. 5 p 5 = (8 , 4) p 6 = (4 , 9) p 7 = (15 , 27) p 8 = (19 , 11) . Is p 5 in ∆( p 1 , p 2 , p 3 ) ? No. 6 Update, and continue... 7

  6. Nonrobustness Real Root Isolation p 7 p 2 Algebraic form of orient p 3 Sign of (( q x − p x )( r y − p y ) − ( q y − p y )( r x − p x )) . ” is + 1, “R” is − 1, “S” is 0. “L p 8 In practice, fl_orient ( p , q , r ) . p 6 p 1 Is p 4 in ∆( p 1 , p 2 , p 3 ) ? 1 p 5 Is fl_orient ( p 1 , p 2 , p 4 ) = L ? Yes. p 4 2 Is fl_orient ( p 2 , p 3 , p 4 ) = L ? Yes. 3 p 1 = (7 . 3000000000000194 , 7 . 3000000000000167) p 2 = (24 . 000000000000068 , 24 . 000000000000071) Is fl_orient ( p 3 , p 1 , p 4 ) = L ? 4 p 3 = (24 . 00000000000005 , 24 . 000000000000053) p 4 = (0 . 50000000000001621 , 0 . 50000000000001243) Yes. 5 p 5 = (8 , 4) p 6 = (4 , 9) p 7 = (15 , 27) p 8 = (19 , 11) . Is p 5 in ∆( p 1 , p 2 , p 3 ) ? No. 6 Update, and continue... 7 p 1 , p 2 , p 3 , p 4 are almost collinear.

  7. Nonrobustness Real Root Isolation Geometry of fl_orient p = ( 0 . 5 , 0 . 5 ) , q = ( 12 , 12 ) , r = ( 24 , 24 ) . fl_orient (( p x + iu , p y + ju ) , q , r ) , 0 ≤ i , j ≤ 255. u = 2 − 53 . Left turn Right turn Collinear Kettner et al.’06 Classroom Examples of Robustness Problems in Geom. Comp.

  8. Nonrobustness Real Root Isolation The Exact Geometric Computation (EGC) Approach

  9. Nonrobustness Real Root Isolation The Exact Geometric Computation (EGC) Approach Geometric Algorithms [Yap’94] Combinatorial structure representing discrete relations amongst geometric objects. E.g., a point is to the left, right or on a line.

  10. Nonrobustness Real Root Isolation The Exact Geometric Computation (EGC) Approach Geometric Algorithms [Yap’94] Combinatorial structure representing discrete relations amongst geometric objects. E.g., a point is to the left, right or on a line. Numerical representation of the geometric objects. E.g., floating-point representation of coordinates.

  11. Nonrobustness Real Root Isolation The Exact Geometric Computation (EGC) Approach Geometric Algorithms [Yap’94] Combinatorial structure representing discrete relations amongst geometric objects. E.g., a point is to the left, right or on a line. Numerical representation of the geometric objects. E.g., floating-point representation of coordinates. Characterize the combinatorial structure by verifying the discrete relations using numerical computations. E.g., using the orientation predicate.

  12. Nonrobustness Real Root Isolation The Exact Geometric Computation (EGC) Approach Geometric Algorithms [Yap’94] Combinatorial structure representing discrete relations amongst geometric objects. E.g., a point is to the left, right or on a line. Numerical representation of the geometric objects. E.g., floating-point representation of coordinates. Characterize the combinatorial structure by verifying the discrete relations using numerical computations. E.g., using the orientation predicate. Cause of Non-robustness Numerical errors may give incorrect characterization.

  13. Nonrobustness Real Root Isolation The Exact Geometric Computation (EGC) Approach The EGC solution Compute correct discrete relations between geometric objects. Correct sign evaluation of geometric predicates. What are geometric predicates?

  14. Nonrobustness Real Root Isolation The Exact Geometric Computation (EGC) Approach q r p Orientation predicate Whether r is to the left, right, or collinear with ( p , q ) ?

  15. Nonrobustness Real Root Isolation The Exact Geometric Computation (EGC) Approach A ( X, Y ) B ( X, Y ) Whether two algebraic curves intersect?

  16. Nonrobustness Real Root Isolation The Exact Geometric Computation (EGC) Approach √ √ 10000134 1000084 √ √ 1000031 √ 10000158 1000087 √ √ 1000182 1000025 √ 1000001 √ 1000198 p q √ √ 1000002 1000199 √ √ 1000175 1000018 √ √ √ 1000113 1000169 1000042 √ √ 1000066 1000116 Which is the shorter path? Almost equal – difference is smaller than 10 − 36 .

  17. Nonrobustness Real Root Isolation The Exact Geometric Computation (EGC) Approach Geometric Predicates Sign of (multivariate) polynomials evaluated at real algebraic numbers Real Algebraic Numbers Real roots of integer polynomials in one variable. √ E.g. ±√ n , for positive integer n , 1 + 5 , but not π , e . 2

  18. Nonrobustness Real Root Isolation The Exact Geometric Computation (EGC) Approach / Numerical Representation of α Data Structure – Rooted DAG, G( α ). + − √ ∗ / k

  19. Nonrobustness Real Root Isolation The Exact Geometric Computation (EGC) Approach / Numerical Representation of α Data Structure – Rooted DAG, G( α ). + − Internal nodes – algebraic operations. √ ∗ / k

  20. Nonrobustness Real Root Isolation The Exact Geometric Computation (EGC) Approach / Numerical Representation of α Data Structure – Rooted DAG, G( α ). + − Internal nodes – algebraic operations. Leaves – integers or real algebraic √ ∗ / k numbers. √ 1+ 5 4 12 23 5 2

  21. Nonrobustness Real Root Isolation The Exact Geometric Computation (EGC) Approach / Numerical Representation of α Data Structure – Rooted DAG, G( α ). + − Internal nodes – algebraic operations. Leaves – integers or real algebraic √ ∗ / k numbers. Isolating interval representation: √ 1+ 5 4 12 23 5 2 A ( X ) ∈ Z [ X ] , A ( α ) = 0, 1 Interval separating α from other roots 2 (conjugates) of A ( X ) .

  22. Nonrobustness Real Root Isolation The Exact Geometric Computation (EGC) Approach / Numerical Representation of α Data Structure – Rooted DAG, G( α ). + − Internal nodes – algebraic operations. Leaves – integers or real algebraic √ ∗ / k numbers. Isolating interval representation: √ 1+ 5 4 12 23 5 2 A ( X ) ∈ Z [ X ] , A ( α ) = 0, 1 Interval separating α from other roots 2 (conjugates) of A ( X ) . Arithmetic is manipulating DAGs.

  23. Nonrobustness Real Root Isolation The Exact Geometric Computation (EGC) Approach / Comparing two numbers: α = β ? Input G ( α ) and G ( β ) . + − √ ∗ / k √ 1+ 5 4 12 23 5 2

  24. Nonrobustness Real Root Isolation The Exact Geometric Computation (EGC) Approach / Comparing two numbers: α = β ? Input G ( α ) and G ( β ) . + − Constructive Zero Bounds : G ( α ) , constructs b s.t. | α | > 2 − b if α � = 0. √ ∗ / k √ 1+ 5 4 12 23 5 2

  25. Nonrobustness Real Root Isolation The Exact Geometric Computation (EGC) Approach / Comparing two numbers: α = β ? Input G ( α ) and G ( β ) . + − Constructive Zero Bounds : G ( α ) , constructs b s.t. | α | > 2 − b if α � = 0. √ ∗ / k √ 1+ 5 4 12 23 5 2 Recursive rules, e.g. | a / b + c / d | ≥ 1 / | cd | , if a > 2 − b then √ a | > 2 − bk . | k Burnikel et al.’99, Li-Yap’00 etc.

  26. Nonrobustness Real Root Isolation The Exact Geometric Computation (EGC) Approach / Comparing two numbers: α = β ? Input G ( α ) and G ( β ) . + − Constructive Zero Bounds : G ( α ) , constructs b s.t. | α | > 2 − b if α � = 0. √ ∗ / k √ 1+ 5 4 12 23 5 2 Recursive rules, e.g. | a / b + c / d | ≥ 1 / | cd | , if a > 2 − b then √ a | > 2 − bk . | k Burnikel et al.’99, Li-Yap’00 etc.

  27. Nonrobustness Real Root Isolation The Exact Geometric Computation (EGC) Approach / Comparing two numbers: α = β ? Input G ( α ) and G ( β ) . + − Constructive Zero Bounds : G ( α ) , constructs b s.t. | α | > 2 − b if α � = 0. √ ∗ / k √ 1+ 5 4 12 23 5 2 Recursive rules, e.g. | a / b + c / d | ≥ 1 / | cd | , if a > 2 − b then √ a | > 2 − bk . | k Burnikel et al.’99, Li-Yap’00 etc.

  28. Nonrobustness Real Root Isolation The Exact Geometric Computation (EGC) Approach | α | > 2 − b / Comparing two numbers: α = β ? Input G ( α ) and G ( β ) . + − Constructive Zero Bounds : G ( α ) , constructs b s.t. | α | > 2 − b if α � = 0. √ ∗ / k √ 1+ 5 4 12 23 5 2 Recursive rules, e.g. | a / b + c / d | ≥ 1 / | cd | , if a > 2 − b then √ a | > 2 − bk . | k Burnikel et al.’99, Li-Yap’00 etc.

  29. Nonrobustness Real Root Isolation The Exact Geometric Computation (EGC) Approach / Comparing two numbers: α = β ? Input G ( α ) and G ( β ) . + − Constructive Zero Bounds : G ( α ) , constructs b s.t. | α | > 2 − b if α � = 0. √ ∗ / k Construct zero bound for G ( α ) ⊖ G ( β ) to get b s.t. | α − β | > 2 − b , if α � = β . √ 1+ 5 4 12 23 5 2 Recursive rules, e.g. | a / b + c / d | ≥ 1 / | cd | , if a > 2 − b then √ a | > 2 − bk . | k Burnikel et al.’99, Li-Yap’00 etc.

  30. Nonrobustness Real Root Isolation The Exact Geometric Computation (EGC) Approach / Comparing two numbers: α = β ? Input G ( α ) and G ( β ) . + − Constructive Zero Bounds : G ( α ) , constructs b s.t. | α | > 2 − b if α � = 0. √ ∗ / k Construct zero bound for G ( α ) ⊖ G ( β ) to get b s.t. | α − β | > 2 − b , if α � = β . √ 1+ 5 4 12 23 5 2 α , � ( b + 1 ) -bit approximations � β ∈ Q . Recursive rules, e.g. | a / b + c / d | ≥ 1 / | cd | , if a > 2 − b then √ a | > 2 − bk . | k Burnikel et al.’99, Li-Yap’00 etc.

  31. Nonrobustness Real Root Isolation The Exact Geometric Computation (EGC) Approach / Comparing two numbers: α = β ? Input G ( α ) and G ( β ) . + − Constructive Zero Bounds : G ( α ) , constructs b s.t. | α | > 2 − b if α � = 0. √ ∗ / k Construct zero bound for G ( α ) ⊖ G ( β ) to get b s.t. | α − β | > 2 − b , if α � = β . √ 1+ 5 4 12 23 5 2 α , � ( b + 1 ) -bit approximations � β ∈ Q . Recursive rules, e.g. α − � β | < 2 − b − 1 then α = β . If | � | a / b + c / d | ≥ 1 / | cd | , if a > 2 − b then √ a | > 2 − bk . | k Burnikel et al.’99, Li-Yap’00 etc.

  32. Nonrobustness Real Root Isolation The Exact Geometric Computation (EGC) Approach / Numerical Representation of α Data Structure – Rooted DAG, G( α ). + − Internal nodes – algebraic 1 operations. √ Leaves – integers or real algebraic 2 ∗ / k numbers. Isolating interval representation: √ 1+ 5 4 12 23 5 2 A ( X ) ∈ Z [ X ] , A ( α ) = 0, 1 Interval separating α from other Recursive rules, e.g. 2 roots of A ( X ) . | a / b + c / d | ≥ 1 / | cd | , Arithmetic is manipulating DAGs. if a > 2 − b then √ Constructive Zero Bounds: G ( α ) , a | > 2 − bk . | k constructs b s.t. | α | > 2 − b if α � = 0. Burnikel et al.’99, Li-Yap’00 etc.

  33. Nonrobustness Real Root Isolation The Exact Geometric Computation (EGC) Approach / Numerical Representation of α Data Structure – Rooted DAG, G( α ). + − Internal nodes – algebraic 1 operations. √ Leaves – integers or real algebraic 2 ∗ / k numbers. Isolating interval representation: √ 1+ 5 4 12 23 5 2 A ( X ) ∈ Z [ X ] , A ( α ) = 0, 1 Interval separating α from other Recursive rules, e.g. 2 roots of A ( X ) . | a / b + c / d | ≥ 1 / | cd | , Arithmetic is manipulating DAGs. if a > 2 − b then √ Constructive Zero Bounds: G ( α ) , a | > 2 − bk . | k constructs b s.t. | α | > 2 − b if α � = 0. Burnikel et al.’99, Li-Yap’00 etc.

  34. Nonrobustness Real Root Isolation The Exact Geometric Computation (EGC) Approach / Numerical Representation of α Data Structure – Rooted DAG, G( α ). + − Internal nodes – algebraic 1 operations. √ Leaves – integers or real algebraic 2 ∗ / k numbers. Isolating interval representation: √ 1+ 5 4 12 23 5 2 A ( X ) ∈ Z [ X ] , A ( α ) = 0, 1 Interval separating α from other Recursive rules, e.g. 2 roots of A ( X ) . | a / b + c / d | ≥ 1 / | cd | , Arithmetic is manipulating DAGs. if a > 2 − b then √ Constructive Zero Bounds: G ( α ) , a | > 2 − bk . | k constructs b s.t. | α | > 2 − b if α � = 0. Burnikel et al.’99, Li-Yap’00 etc.

  35. Nonrobustness Real Root Isolation The Exact Geometric Computation Approach Implementations Leda reals – http://www.mpi-inf.mpg.de/LEDA/ Core Library – http://www.cs.nyu.edu/exact/core Used in CGAL – www.cgal.org

  36. Nonrobustness Real Root Isolation Theoretical Foundations of EGC Ideal World Algorithms in Real RAM model. Computes a function f : R → R .

  37. Nonrobustness Real Root Isolation Theoretical Foundations of EGC Ideal World Algorithms in Real RAM model. Computes a function f : R → R . The EGC World, [Yap, 2003] � f is f implemented in EGC model. Input to � f is a dense subset (say Q ) of R and precision p . For x ∈ Q , � f computes a relative approx. to f i.e. � f ( x ) = f ( x )( 1 ± 2 − p ) . What class of functions are EGC-computable?

  38. Nonrobustness Real Root Isolation Theoretical Foundations of EGC Ideal World Algorithms in Real RAM model. Computes a function f : R → R . The EGC World, [Yap, 2003] � f is f implemented in EGC model. Input to � f is a dense subset (say Q ) of R and precision p . For x ∈ Q , � f computes a relative approx. to f i.e. � f ( x ) = f ( x )( 1 ± 2 − p ) . What class of functions are EGC-computable?

  39. Nonrobustness Real Root Isolation Fundamental Problem Zero Problem Are two real numbers α , β equal?

  40. Nonrobustness Real Root Isolation Fundamental Problem Zero Problem Are two real numbers α , β equal? / Numerical Representation of α Data Structure – Rooted DAG, G( α ). + − Isolating interval representation: A ( X ) ∈ Z [ X ] , A ( α ) = 0, 1 √ ∗ Interval separating α from other / k 2 roots of A ( X ) . √ Constructive Zero Bounds: G ( α ) , 1+ 5 4 12 23 5 2 constructs b s.t. | α | > 2 − b if α � = 0.

  41. Nonrobustness Real Root Isolation Fundamental Problem Zero Problem Are two real numbers α , β equal? / Numerical Representation of α Data Structure – Rooted DAG, G( α ). + − Isolating interval representation: A ( X ) ∈ Z [ X ] , A ( α ) = 0, 1 √ ∗ Interval separating α from other / k 2 roots of A ( X ) . √ Constructive Zero Bounds: G ( α ) , 1+ 5 4 12 23 5 2 constructs b s.t. | α | > 2 − b if α � = 0. Transcendental Functions Interior nodes were exp, log, sin, cos etc., or leaves e , π ?

  42. Nonrobustness Real Root Isolation Fundamental Problem Zero Problem Are two real numbers α , β equal? / Numerical Representation of α Data Structure – Rooted DAG, G( α ). + − Isolating interval representation: A ( X ) ∈ Z [ X ] , A ( α ) = 0, 1 √ ∗ Interval separating α from other / k 2 roots of A ( X ) . √ Constructive Zero Bounds: G ( α ) , 1+ 5 4 12 23 5 2 constructs b s.t. | α | > 2 − b if α � = 0. Transcendental Functions Interior nodes were exp, log, sin, cos etc., or leaves e , π ?

  43. Nonrobustness Real Root Isolation Zero Problem for exp, log Expressions, Richardson’07 exp, log Expressions √}∪{ exp , log } Interior nodes are { + , − , ∗ ,/, k

  44. Nonrobustness Real Root Isolation Zero Problem for exp, log Expressions, Richardson’07 exp, log Expressions √}∪{ exp , log } Interior nodes are { + , − , ∗ ,/, k Schanuel’s Conjecture Given n complex numbers z 1 ,..., z n linearly independent over Q . At least n transcendental numbers in { z 1 ,..., z n , e z 1 ,..., e z n } . Generalizes Lindemann-Weierstrass Theorem z 1 ,..., z n are n linearly independent algebraic numbers.

  45. Nonrobustness Real Root Isolation Complexity of Algebraic Numbers Which algebraic numbers can be relatively approximated in poly time? √ √ Sum of Square Roots 10000134 1000084 √ √ √ 1000031 √ √ a i − ∑ k 10000158 1000087 ∑ k b i , i = 1 i = 1 √ √ 1000025 1000182 | a i | , | b i | ≤ N . √ 1000001 √ S ( N , k ) , the minimum 1000198 p q positive absolute value √ √ 1000002 of the sum. 1000199 √ √ Current bounds: 1000175 1000018 S ( N , k ) � N − 2 k . √ √ √ 1000113 1000169 1000042 √ √ 1000066 1000116 Hope: S ( N , k ) � N − k .

  46. Nonrobustness Real Root Isolation Real Root Isolation The Problem Given A ( X ) ∈ Z [ X ] , degree d . A ( X )

  47. Nonrobustness Real Root Isolation Real Root Isolation The Problem Given A ( X ) ∈ Z [ X ] , degree d . A ( X )

  48. Nonrobustness Real Root Isolation Real Root Isolation The Problem Given A ( X ) ∈ Z [ X ] , degree d . A ( X )

  49. Nonrobustness Real Root Isolation Real Root Isolation The Problem Given A ( X ) ∈ Z [ X ] , degree d . A ( X ) Assumption All the roots are of multiplicity one, i.e. GCD ( A ( X ) , A ′ ( X )) = 1.

  50. Nonrobustness Real Root Isolation Selective History Classical Work Descartes, Newton, Fourier, Sturm, Vincent, Weierstrass, Obreshkoff, Ostrowski, Weyl, Henrici, ... Modern Work – Complexity and Implementation Schönhage, Smale, Pan (optimal complexity results), ... Collins, Johnson, Krandick, Bini, Mehlhorn, Sagraloff, ... Relevance Fundamental problem in computational algebra, used in Cylindrical Algebraic Decomposition, in Ray Tracing, Computer Aided Design, for verifying conjectures, ...

  51. Nonrobustness Real Root Isolation A General Subdivision Algorithm Input: A ( X ) ∈ Z [ X ] of degree d , and I 0 . Output: Isolating intervals for roots of A ( X ) in I 0 .

  52. Nonrobustness Real Root Isolation A General Subdivision Algorithm Input: A ( X ) ∈ Z [ X ] of degree d , and I 0 . Output: Isolating intervals for roots of A ( X ) in I 0 . The Real Root Counting function Count ( A , I ) = number of real roots of A ( X ) in I .

  53. Nonrobustness Real Root Isolation A General Subdivision Algorithm Input: A ( X ) ∈ Z [ X ] of degree d , and I 0 . Output: Isolating intervals for roots of A ( X ) in I 0 . The Real Root Counting function Count ( A , I ) = number of real roots of A ( X ) in I . RootIsol( A ( X ) , I ) If Count ( A , I ) = 0 then return. 1 If Count ( A , I ) = 1 then output I and return. 2 Let m be the midpoint of I = ( I l , I r ) . 3 Recurse on ( A ( X ) , ( I l , m )) and ( A ( X ) , ( m , I r )) . 4

  54. Nonrobustness Real Root Isolation Illustration I 0 7

  55. Nonrobustness Real Root Isolation Illustration I 0 7 4 3

  56. Nonrobustness Real Root Isolation Illustration I 0 7 4 3 0 4 1 2

  57. Nonrobustness Real Root Isolation Illustration I 0 7 4 3 0 4 1 2

  58. Nonrobustness Real Root Isolation Sturm Sequences The Sequence A := ( A 0 = A , A 1 = A ′ ,..., A k ) , A i + 1 := rem ( A i − 1 , A i ) .

  59. Nonrobustness Real Root Isolation Sturm Sequences The Sequence A := ( A 0 = A , A 1 = A ′ ,..., A k ) , A i + 1 := − rem ( A i − 1 , A i ) .

  60. Nonrobustness Real Root Isolation Sturm Sequences The Sequence A := ( A 0 = A , A 1 = A ′ ,..., A k ) , A i + 1 := − rem ( A i − 1 , A i ) . A ( x ) = ( x 2 − x − 1 ) , A ′ ( x ) = 2 x − 1 x 2 − x − 1 = ( 2 x − 1 ) ( 2 x − 1 ) 4 = 4 x 2 − 4 x + 1 − 5 − 5 4 . 4 4 A = ( x 2 − x − 1 , 2 x − 1 , − 5 4 ) .

  61. Nonrobustness Real Root Isolation Sturm Sequences The Sequence A := ( A 0 = A , A 1 = A ′ ,..., A k ) , A i + 1 := − rem ( A i − 1 , A i ) . A ( x ) = ( x 2 − x − 1 ) , A ′ ( x ) = 2 x − 1 x 2 − x − 1 = ( 2 x − 1 ) ( 2 x − 1 ) 4 = 4 x 2 − 4 x + 1 − 5 − 5 4 . 4 4 A = ( x 2 − x − 1 , 2 x − 1 , + 5 4 ) .

  62. Nonrobustness Real Root Isolation Sturm Sequences The Sequence A := ( A 0 = A , A 1 = A ′ ,..., A k ) , A i + 1 := − rem ( A i − 1 , A i ) . A ( x ) = ( x 2 − x − 1 ) , A ′ ( x ) = 2 x − 1 x 2 − x − 1 = ( 2 x − 1 ) ( 2 x − 1 ) 4 = 4 x 2 − 4 x + 1 − 5 − 5 4 . 4 4 A = ( x 2 − x − 1 , 2 x − 1 , + 5 4 ) . The Variation Given c ∈ R , evaluate A at c , i.e., ( A 0 ( c ) , A 1 ( c ) ,..., A k ( c )) . Drop all the zeros from the sequence ( A 0 ( c ) , A 1 ( c ) ,..., A k ( c )) . Var ( A ; c ) := no. of sign flips from + to − or vice versa.

  63. Nonrobustness Real Root Isolation Sturm Sequences The Sequence A := ( A 0 = A , A 1 = A ′ ,..., A k ) , A i + 1 := − rem ( A i − 1 , A i ) . A ( x ) = ( x 2 − x − 1 ) , A ′ ( x ) = 2 x − 1 x 2 − x − 1 = ( 2 x − 1 ) ( 2 x − 1 ) 4 = 4 x 2 − 4 x + 1 − 5 − 5 4 . 4 4 A = ( x 2 − x − 1 , 2 x − 1 , + 5 4 ) . Var ( A ; 1 ) = #( − , + , +) = 1, Var ( A ; 2 ) = #(+ , + , +) = 0. The Variation Given c ∈ R , evaluate A at c , i.e., ( A 0 ( c ) , A 1 ( c ) ,..., A k ( c )) . Drop all the zeros from the sequence ( A 0 ( c ) , A 1 ( c ) ,..., A k ( c )) . Var ( A ; c ) := no. of sign flips from + to − or vice versa.

  64. Nonrobustness Real Root Isolation Sturm Sequences The Sequence A := ( A 0 = A , A 1 = A ′ ,..., A k ) , A i + 1 := − rem ( A i − 1 , A i ) . A ( x ) = ( x 2 − x − 1 ) , A ′ ( x ) = 2 x − 1 x 2 − x − 1 = ( 2 x − 1 ) ( 2 x − 1 ) 4 = 4 x 2 − 4 x + 1 − 5 − 5 4 . 4 4 A = ( x 2 − x − 1 , 2 x − 1 , + 5 4 ) . Var ( A ; 1 ) = #( − , + , +) = 1, Var ( A ; 2 ) = #(+ , + , +) = 0. The Variation Given c ∈ R , evaluate A at c , i.e., ( A 0 ( c ) , A 1 ( c ) ,..., A k ( c )) . Drop all the zeros from the sequence ( A 0 ( c ) , A 1 ( c ) ,..., A k ( c )) . Var ( A ; c ) := no. of sign flips from + to − or vice versa. Sturm’s Theorem, 1829 No. of real roots of A ( x ) in ( c , d ) = Var ( A ; c ) − Var ( A ; d ) .

  65. Nonrobustness Real Root Isolation Complexity Analysis – Sturm’s Algorithm, Davenport’85 I 0 7 4 3 0 4 1 2 3 1 1 1 1 2 1 1 Size of the subdivision tree, | T | . 1 Worst case complexity at every node. 2

  66. Nonrobustness Real Root Isolation Complexity Analysis – Sturm’s Algorithm, Davenport’85 I 0 7 4 3 0 4 1 2 3 1 1 1 1 2 1 1 Size of the subdivision tree, | T | . 1 Worst case complexity at every node. 2

  67. Nonrobustness Real Root Isolation Complexity Analysis – Sturm’s Algorithm Measure of Complexity Root Separation of A , sep ( A ) := min {| α − β | : α , β ∈ Z ( A ) ⊆ C } .

  68. Nonrobustness Real Root Isolation Complexity Analysis – Sturm’s Algorithm Measure of Complexity Root Separation of A , sep ( A ) := min {| α − β | : α , β ∈ Z ( A ) ⊆ C } . Bounds i = 0 a i x i ∈ Z [ x ] , degree d , | a i | ≤ 2 L , i = 0 ,..., d . A ( x ) = ∑ d − logsep ( A ) = O ( dL + d log d ) . w ( I 0 ) < 2 L .

  69. Nonrobustness Real Root Isolation Complexity Analysis – Sturm’s Algorithm I 0 7 4 3 0 4 1 2 3 1 1 1 1 2 1 1

  70. Nonrobustness Real Root Isolation Complexity Analysis – Sturm’s Algorithm I 0 7 T ′ 4 3 0 4 1 2 3 1 1 1 1 2 1 1

  71. Nonrobustness Real Root Isolation Complexity Analysis – Sturm’s Algorithm I 0 7 | T | ≤ 2 | T ′ | 4 3 0 4 1 2 3 1 1 1 1 2 1 1

  72. Nonrobustness Real Root Isolation Complexity Analysis – Sturm’s Algorithm I 0 7 4 3 4 2 3 2

  73. Nonrobustness Real Root Isolation Complexity Analysis – Sturm’s Algorithm I 0 7 4 3 J 4 2 3 2 α J β J

  74. Nonrobustness Real Root Isolation Complexity Analysis – Sturm’s Algorithm I 0 7 4 3 J 4 2 w ( I 0 ) log 3 | α J − β J | 2 α J β J

  75. Nonrobustness Real Root Isolation Complexity Analysis – Sturm’s Algorithm I 0 7 4 3 J 4 2 w ( I 0 ) log 3 | α J − β J | 2 α J β J

  76. Nonrobustness Real Root Isolation Complexity Analysis – Sturm’s Algorithm I 0 7 4 3 J 4 2 w ( I 0 ) log 3 | α J − β J | 2 α J β J w ( I 0 ) | T ′ | ≤ ∑ J log | α J − β J |

  77. Nonrobustness Real Root Isolation Complexity Analysis – Sturm’s Algorithm I 0 7 4 3 J 4 2 w ( I 0 ) log 3 | α J − β J | 2 α J β J w ( I 0 ) | T ′ | ≤ ∑ J log | α J − β J | = O ( d log w ( I 0 )) − ∑ J log | α J − β J |

  78. Nonrobustness Real Root Isolation Complexity Analysis – Sturm’s Algorithm I 0 7 4 3 J 4 2 w ( I 0 ) log 3 | α J − β J | 2 α J β J w ( I 0 ) | T ′ | ≤ ∑ J log | α J − β J | = O ( d log w ( I 0 )) − ∑ J log | α J − β J | − ∑ J log | α J − β J |

  79. Nonrobustness Real Root Isolation Complexity Analysis – Sturm’s Algorithm I 0 7 4 3 J 4 2 w ( I 0 ) log 3 | α J − β J | 2 α J β J w ( I 0 ) | T ′ | ≤ ∑ J log | α J − β J | = O ( d log w ( I 0 )) − ∑ J log | α J − β J | − ∑ J log | α J − β J | = O ( − d logsep ( A )) = O ( d ( dL + d log d ))

  80. Nonrobustness Real Root Isolation Complexity Analysis – Sturm’s Algorithm I 0 7 4 3 J 4 2 w ( I 0 ) log 3 | α J − β J | 2 α J β J w ( I 0 ) | T ′ | ≤ ∑ J log | α J − β J | = O ( d log w ( I 0 )) − ∑ J log | α J − β J | − ∑ J log | α J − β J | = O ( − d logsep ( A )) = O ( d ( dL + d log d )) − ∑ J log | α J − β J | = O ( dL + d log d ) , Davenport-Mahler.

  81. Nonrobustness Real Root Isolation Some Remarks on Sturm’s Algorithm The subdivision tree size is optimal (Mignotte’s polynomials).

Recommend

Numerical Robustness (for Geometric Calculations) Christer Ericson Sony Computer Entertainment

Numerical Robustness (for Geometric Calculations) Christer Ericson Sony Computer Entertainment

Numerical Robustness (for Geometric Calculations) Christer Ericson Sony Computer Entertainment Slides @ http:/ / realtimecollisiondetection.net/ pubs/ Numerical Robustness (for Geometric Calculations) Christer Ericson Sony Computer

704 views • 52 slides
A geometric perspective on the robustness of deep networks Seyed-Mohsen Moosavi-Dezfooli

A geometric perspective on the robustness of deep networks Seyed-Mohsen Moosavi-Dezfooli

A geometric perspective on the robustness of deep networks Seyed-Mohsen Moosavi-Dezfooli Amirkabir Artificial Intelligence Summer Summit July 2019 Tehran Polytechnic Iran 2 EPFL Lausanne, Switzerland 3 Pascal Frossard Alhussein

731 views • 44 slides
Soft Foundations for Geometric Computation Chee Yap Courant Institute, NYU (Visiting) Academy

Soft Foundations for Geometric Computation Chee Yap Courant Institute, NYU (Visiting) Academy

Soft Foundations for Geometric Computation Chee Yap Courant Institute, NYU (Visiting) Academy of Mathematics &amp; System Sciences Chinese Academy of Scieces, Beijing Geometric Computation and Applications Hamilton Mathematics Institute

497 views • 37 slides
Level set methods for robustness measures P. V AN D OOREN CESAME, Universit e catholique de

Level set methods for robustness measures P. V AN D OOREN CESAME, Universit e catholique de

Level set methods for robustness measures P. V AN D OOREN CESAME, Universit e catholique de Louvain joint work with Y. Genin, T. Lawrence, M. Overton, J. Sreedhar, A. Tits What it is about Fast and reliable computation of robustness measures

490 views • 29 slides
Robustness measures and level set methods P. Van Dooren Abstract In this talk we discuss the

Robustness measures and level set methods P. Van Dooren Abstract In this talk we discuss the

Robustness measures and level set methods P. Van Dooren Abstract In this talk we discuss the computation of a range of robustness measures for linear time invariant systems. More precisely, we propose algorithms based on the idea of level set

524 views • 5 slides
Robustness? Robustness ? Robustness?

Robustness? Robustness ? Robustness?

Robustness? Robustness ? Robustness? Thomas Mandl

245 views • 5 slides
Where Are We? Lecture 9 Robustness through Training 1 Robustness Explicit Handling of Noise

Where Are We? Lecture 9 Robustness through Training 1 Robustness Explicit Handling of Noise

Where Are We? Lecture 9 Robustness through Training 1 Robustness Explicit Handling of Noise and Channel Variations 2 Michael Picheny, Bhuvana Ramabhadran, Stanley F. Chen Robustness via Adaptation 3 IBM T.J. Watson Research Center

256 views • 22 slides
A Geometric Approach for the Computation of Riemann-Roch Spaces : Algorithm and Complexity Aude Le

A Geometric Approach for the Computation of Riemann-Roch Spaces : Algorithm and Complexity Aude Le

A Geometric Approach for the Computation of Riemann-Roch Spaces : Algorithm and Complexity Aude Le Gluher and Pierre-Jean Spaenlehauer Universit de Lorraine / INRIA Nancy Grand Est / CNRS CARAMBA team JNCF, Luminy, 2019 1 / 26

484 views • 29 slides
Click to edit Master title style Skeleton computation of an image using a geometric approach J.

Click to edit Master title style Skeleton computation of an image using a geometric approach J.

Click to edit Master title style Skeleton computation of an image using a geometric approach J. Martinez, M. Vigo, N. Pla-Garcia, D. Ayala Introduction One of the challenges of BioCAD field is to understand the morphology of the pore

412 views • 29 slides
UCSD Robustness Summer School David Donoho 20190812 David Donoho UCSD Robustness Summer School

UCSD Robustness Summer School David Donoho 20190812 David Donoho UCSD Robustness Summer School

What is Statistics? What is Statistical Research? Paradigm Shifts in Statistics How did Robustness in Statistics Emerge? What has Robustness in Statistics delivered? What has Robustness in Statistics to do with CS? UCSD Robustness

721 views • 26 slides
Robustness and Generalization Huan Xu The University of Texas at Austin Department of Electrical

Robustness and Generalization Huan Xu The University of Texas at Austin Department of Electrical

Robustness and Generalization Huan Xu The University of Texas at Austin Department of Electrical and Computer Engineering COLT, June 29, 2010 Joint work with Shie Mannor What is Robustness? What is Robustness? What is Robustness?

571 views • 53 slides
Robustness and SMC Adam Pechner Overview What is Robustness and why do we care? Different

Robustness and SMC Adam Pechner Overview What is Robustness and why do we care? Different

Robustness and SMC Adam Pechner Overview What is Robustness and why do we care? Different types of Robust Control Techniques Sliding Mode Control (SMC) Definition and Benefits Drawbacks and Requirements Applications of SMC

538 views • 40 slides
Chapter 8 Binomial and Geometric Distribu7ons 8.2 Geometric

Chapter 8 Binomial and Geometric Distribu7ons 8.2 Geometric

12/9/12 Chapter 8 Binomial and Geometric Distribu7ons 8.2 Geometric Distributions Waiting for Sammy Sosa geometric distribution Children s cereals sometimes contain prizes. Imagine that packages of

376 views • 9 slides
Algorithms in Nature Network robustness Slides adapted from Carl Kingsford Network robustness

Algorithms in Nature Network robustness Slides adapted from Carl Kingsford Network robustness

Algorithms in Nature Network robustness Slides adapted from Carl Kingsford Network robustness Many complex systems show a surprising degree of tolerance to errors: Biological networks persist despite environmental noise, failures, attacks

224 views • 20 slides
Point sets, Maps and Navigation - II D.A. Forsyth Robustness is a serious problem Robustness is

Point sets, Maps and Navigation - II D.A. Forsyth Robustness is a serious problem Robustness is

Point sets, Maps and Navigation - II D.A. Forsyth Robustness is a serious problem Robustness is a serious problem Key issue: Squaring a large number produces a huge number A few wildly mismatch points can throw off R, t Fixes:

153 views • 12 slides
Geometric Optimization Piotr Indyk April 26, 2005 Lecture 19: Geometric Optimization Geometric

Geometric Optimization Piotr Indyk April 26, 2005 Lecture 19: Geometric Optimization Geometric

Geometric Optimization Piotr Indyk April 26, 2005 Lecture 19: Geometric Optimization Geometric Optimization Minimize/maximize something subject to some constraints Have seen: Linear Programming Minimum Enclosing Ball

337 views • 17 slides
Geometric Representations 3D Graphics Motivation Geometric representation What do we want

Geometric Representations 3D Graphics Motivation Geometric representation What do we want

Geometric Representations 3D Graphics Motivation Geometric representation What do we want to do? empty space (typically 3 ) B geometric object B 3 Fundamental problem The Problem: d B infinite number of

954 views • 47 slides
2D Geometric Transformations Question : How do we represent a geometric object in the plane?

2D Geometric Transformations Question : How do we represent a geometric object in the plane?

2D Geometric Transformations (Chapter 5 in FVD) 1 2D Geometric Transformations Question : How do we represent a geometric object in the plane? Answer : For now, assume that objects consist of points and lines. A point is represented

353 views • 13 slides
Trade-off between Efficiency and Robustness Doctoral Colloqium @ SenSys18, Shenzhen Robert

Trade-off between Efficiency and Robustness Doctoral Colloqium @ SenSys18, Shenzhen Robert

Institute of Operating Systems and Computer Networks E ffi ciency E ffi ciency vs Robustness = Trade-o ff Robustness Energy Budget Trade-off between Efficiency and Robustness Doctoral Colloqium @ SenSys18, Shenzhen Robert Hartung,

680 views • 19 slides
Geometric Modeling An Example 2 Geomertic Modeling I - Center for Graphics and Geometric

Geometric Modeling An Example 2 Geomertic Modeling I - Center for Graphics and Geometric

Geometric Modeling An Example 2 Geomertic Modeling I - Center for Graphics and Geometric Computing, Technion Outline Objective: Develop methods and algorithms to mathematically model shapes of real world objects Categories:

975 views • 59 slides
S9932: LEARNING TO BOOST S9932: LEARNING TO BOOST ROBUSTNESS FOR ROBUSTNESS FOR AUTONOMOUS

S9932: LEARNING TO BOOST S9932: LEARNING TO BOOST ROBUSTNESS FOR ROBUSTNESS FOR AUTONOMOUS

S9932: LEARNING TO BOOST S9932: LEARNING TO BOOST ROBUSTNESS FOR ROBUSTNESS FOR AUTONOMOUS DRIVING AUTONOMOUS DRIVING Bernhard Firner, March 20, 2019 AUTONOMOUS DRIVING Sounds simple Actually pretty diffjcult Can start with sub-domains to

798 views • 49 slides
Key-Robustness for Cryptographic Primitives ie 1 R azvan Ros 1 ENS, CNRS, INRIA &amp; PSL

Key-Robustness for Cryptographic Primitives ie 1 R azvan Ros 1 ENS, CNRS, INRIA &amp; PSL

Key-Robustness for Cryptographic Primitives ie 1 R azvan Ros 1 ENS, CNRS, INRIA &amp; PSL Research University, Paris, France ECRYPT-NET Summer School, Crete, Greece 12 th October 2017 R azvan Ros ie Key-Robustness for Cryptographic

949 views • 48 slides
Digital/Analog Computation in the Cell Computational Systems Biology and Optimization Fran

Digital/Analog Computation in the Cell Computational Systems Biology and Optimization Fran

FO-LTL(R) Continuous Satisfaction Degree Robustness Compiler Ptime Analog Digital/Analog Computation in the Cell Computational Systems Biology and Optimization Fran cois Fages Lifeware group Inria Saclay 1 / 9 FO-LTL(R) Continuous

440 views • 9 slides
Matrix Robustness, with an Application to Power System Observability Matthias Brosemann Jochen

Matrix Robustness, with an Application to Power System Observability Matthias Brosemann Jochen

Power system observability Complexity of Matrix Robustness Algorithms for Matrix Robustness Experiments Matrix Robustness, with an Application to Power System Observability Matthias Brosemann Jochen Alber Falk H uffner Rolf Niedermeier

350 views • 31 slides

More recommend