characterizing algebraic invariants by differential
play

Characterizing Algebraic Invariants by Differential Radical - PowerPoint PPT Presentation

Characterizing Algebraic Invariants by Differential Radical Invariants Khalil Ghorbal Carnegie Mellon university Joint work with Andr e Platzer CMACS AVACS November 21st, 2013 K. Ghorbal (CMU) CMACS AVACS 1 CMACS 1 / 31 Introduction


  1. Characterizing Algebraic Invariants by Differential Radical Invariants Khalil Ghorbal Carnegie Mellon university Joint work with Andr´ e Platzer CMACS AVACS November 21st, 2013 K. Ghorbal (CMU) CMACS AVACS 1 CMACS 1 / 31

  2. Introduction Context: ODE in Computer Science/Formal Verification Goal. Automated Formal Reasoning about Ordinary Differential Equations. Formal Reasoning: Global Properties of All solutions. Applications to the Formal Verification of Hybrid Systems • Reachability Analysis • Proof Rules • Synthesis Useful in many other fields: Control Theory, Stability Analysis, Numerical Integration, Integrability of ODE. K. Ghorbal (CMU) CMACS AVACS 2 CMACS 2 / 31

  3. Introduction Algebraic Differential Equations Example x ι = (1 , 0 , 0 , 1) x 1 = − x 2 ˙ x 2 = x 1 ˙ x 3 = x 2 ˙ x 4 = x 3 x 4 ˙ 4 Formally, we study the Initial Value Problem: dx i ( t ) x i = p i ( x ) , 1 ≤ i ≤ n , x (0) = x ι ∈ R n . = ˙ dt ⊕ Parameters are allowed ⊕ Many analytic functions can be encoded (sin, cos, ln, . . . ) ⊕ / ⊖ The initial value ( x ι ) are not restricted ⊖ Evolution domain abstracted (still sound) K. Ghorbal (CMU) CMACS AVACS 3 CMACS 3 / 31

  4. Introduction Approach Algebraic Invariant Expression ∀ t , h ( x ( t )) = 0 , for all x ( t ) solution of the Initial Value Problem. Tools • Classical Algebraic Geometry: Polynomial Ring, Ideals, Varieties • Symbolic Linear Algebra K. Ghorbal (CMU) CMACS AVACS 4 CMACS 4 / 31

  5. Time Abstraction Outline Introduction 1 Time Abstraction 2 Characterization of Invariant Expressions 3 Automated Generation 4 Conclusion 5 K. Ghorbal (CMU) CMACS AVACS 5 CMACS 5 / 31

  6. Time Abstraction Orbits Definition O ( x ι ) def = { x ( t ) | t ∈ U t } ⊆ R n U t domain of definition of the maximal solution of the Initial Value Problem (˙ x = p ( x ) , x (0) = x ι ). Example Solar System Galileo Orbit K. Ghorbal (CMU) CMACS AVACS 6 CMACS 6 / 31

  7. Time Abstraction Orbits: Issues Example 1.0 1.0 0.5 0.5 � 1.0 � 0.5 0.5 1.0 � 1.0 � 0.5 0.5 1.0 � 0.5 � 0.5 � 1.0 � 1.0 Cornu Spiral Lissajous Curve Solutions → Exact Orbit • Computation issues • Decidability issues Idea: Time Abstraction � K. Ghorbal (CMU) CMACS AVACS 7 CMACS 7 / 31

  8. Time Abstraction Affine Varieties and Ideals Roots of h 2 Polynomials 1 h def = x 4 1 + x 2 2 − 2 0 What about the polynomials ph ? � 1 � 2 � 2 � 1 0 1 2 Ideal : stable set of polynomials under external multiplication I = � h 1 , . . . , h r � def = { � r i =1 g i h i | g 1 , . . . , g r ∈ R [ x ] } Affine Variety : common roots of all polynomials in I = { x ∈ R n | ∀ h ∈ I , h ( x ) = 0 } V ( I ) def K. Ghorbal (CMU) CMACS AVACS 8 CMACS 8 / 31

  9. Time Abstraction Affine Varieties and Ideals Roots of h 2 Polynomials 1 h def = x 4 1 + x 2 2 − 2 0 What about the polynomials ph ? � 1 � 2 � 2 � 1 0 1 2 Ideal : stable set of polynomials under external multiplication I = � h 1 , . . . , h r � def = { � r i =1 g i h i | g 1 , . . . , g r ∈ R [ x ] } Affine Variety : common roots of all polynomials in I = { x ∈ R n | ∀ h ∈ I , h ( x ) = 0 } V ( I ) def K. Ghorbal (CMU) CMACS AVACS 8 CMACS 8 / 31

  10. Time Abstraction Variety Embedding of Orbits Zariski Closure Vanishing Ideal : all polynomials that vanish on O ( x ι ) I ( O ( x ι )) def = { h ∈ R [ x ] | ∀ x ∈ O ( x ι ) , h ( x ) = 0 } Closure: Sound Abstraction O ( x ι ) def O ( x ι ) ⊆ ¯ = V ( I ( O ( x ι ))) Orbit − → Vanishing Ideal − → Closure ⊇ Orbit Closure is the smallest variety that contains Orbit . Example x = x � x ( t ) = x ι e t � O ( x ι ) = [ x ι , ∞ [ � I = � 0 � � ¯ ˙ O ( x ι ) = R K. Ghorbal (CMU) CMACS AVACS 9 CMACS 9 / 31

  11. Time Abstraction Example: Variety Embedding Zariski Closure (Intuition) K. Ghorbal (CMU) CMACS AVACS 10 CMACS 10 / 31

  12. Characterization of Invariant Expressions Outline Introduction 1 Time Abstraction 2 Characterization of Invariant Expressions 3 Automated Generation 4 Conclusion 5 K. Ghorbal (CMU) CMACS AVACS 11 CMACS 11 / 31

  13. Characterization of Invariant Expressions Hold on ... Sound Abstraction Orbit ⊆ Closure Goal Explicit Characterization of the Vanishing Ideal I ( O ( x ι )) K. Ghorbal (CMU) CMACS AVACS 12 CMACS 12 / 31

  14. Characterization of Invariant Expressions Lie Derivation Lie derivative along a vector field n ∂ h L p ( h ) def � = p i ( x ) ∂ x i i =1 Properties • Algebraic differentiation • Applies to the polynomial h (not the function t �→ h ( x ( t ))) • Corresponds to the time derivative when the solution is substituted back The Vanishing Ideal is a Differential Ideal L p ( h ) ∈ I ( O ( x ι )) for all h ∈ I ( O ( x ι )). K. Ghorbal (CMU) CMACS AVACS 13 CMACS 13 / 31

  15. Characterization of Invariant Expressions Differential Radical Invariants Theorem h ∈ I ( O ( x ι )) if and only if there exists a finite integer N s.t. L ( N ) ( h ) ∈ � L (0) p ( h ) , . . . , L ( N − 1) ( h ) � ⊆ I ( O ( x ι )) ( ı ) p p L (0) p ( h )( x ι ) = 0 , . . . , L ( N − 1) ( h )( x ι ) = 0 . ( ıı ) p Proof Sketch “ ⇒ ” Ascending Chain Condition on ideals ( R [ x ] is Notherian ) “ ⇐ ” (Global) Cauchy-Lipschitz Theorem K. Ghorbal (CMU) CMACS AVACS 14 CMACS 14 / 31

  16. Characterization of Invariant Expressions Special Case: Invariant (Algebraic) Functions N = 1 and L p ( h ) ∈ � 0 � • L p ( h ) = 0 ∧ h ( x ι ) = 0 − → h = 0 Example 2 2 2 1 1 1 0 0 0 � 1 � 1 � 1 � 2 � 2 � 2 � 2 � 1 0 1 2 � 2 � 1 0 1 2 � 2 � 1 0 1 2 Vector Field ˙ x 1 = − x 2 , Roots of Roots of L p ( h ): h def = x 2 1 + x 2 x 2 = x 1 , x ι = (1 , 0) ˙ 2 − 1 Whole Space K. Ghorbal (CMU) CMACS AVACS 15 CMACS 15 / 31

  17. Characterization of Invariant Expressions Special Case ( N = 1) Darboux Invariants a.k.a. λ -Invariant, Exponential Invariants, P -Consecution, • L p ( h ) = ph ∧ h ( x ι ) = 0 − → h = 0 x 1 = − x 1 + 2 x 2 h = ( x ι 2 − x ι 1 x ι 2 2 ) x 1 − x ι 1 ( x 2 − x 1 x 2 ˙ 2 ) 1 x 2 x 2 = x 2 ˙ L p ( h ) = ( − 1 + 2 x 1 x 2 ) h Example 10 10 1.0 5 5 0.5 0 0.0 0 � 0.5 � 5 � 5 � 1.0 � 10 � 10 � 10 � 5 0 5 10 � 1.0 � 0.5 0.0 0.5 1.0 � 10 � 5 0 5 10 Roots of L p ( h ) Vector Field Roots of h K. Ghorbal (CMU) CMACS AVACS 16 CMACS 16 / 31

  18. Characterization of Invariant Expressions Decidability Corollary It is decidable whether a polynomial h with real algebraic coefficients is an algebraic invariant of an algebraic differential system with real algebraic coefficients and real algebraic initial values. Related Work Generalizes the decidability of invariant functions [A. Platzer ITP’12] K. Ghorbal (CMU) CMACS AVACS 17 CMACS 17 / 31

  19. Characterization of Invariant Expressions Sound Approximation of the Closure ¯ O ( x ι ) Differential Radical Ideals def = � L ( i ) J j p ( h j ) � 0 ≤ i ≤ N − 1 Underapproximation of I ( O ( x ι )) � J j = I ( O ( x ι )) , ℑ finite j ∈ℑ Overapproximation of ¯ O ( x ι ) ¯ � O ( x ι ) ⊆ V ( J i ) 1 ≤ i ≤ r K. Ghorbal (CMU) CMACS AVACS 18 CMACS 18 / 31

  20. Characterization of Invariant Expressions Example System x 1 = − x 2 ˙ x 2 = x 1 ˙ x 3 = x 2 ˙ x 4 = x 3 x 4 ˙ 4 Differential Radical Invariants h 1 = x 3 − x 2 x 4 and h 2 = x 2 4 − x 2 3 − 1 Roots of h 1 Roots of h 2 Orbit K. Ghorbal (CMU) CMACS AVACS 19 CMACS 19 / 31

  21. Characterization of Invariant Expressions Example: cont’d Overapproximation of ¯ O ( x ι ) K. Ghorbal (CMU) CMACS AVACS 20 CMACS 20 / 31

  22. Automated Generation Outline Introduction 1 Time Abstraction 2 Characterization of Invariant Expressions 3 Automated Generation 4 Conclusion 5 K. Ghorbal (CMU) CMACS AVACS 21 CMACS 21 / 31

  23. Automated Generation So ... Sound Abstraction Orbit ⊆ Closure Characterization of I ( O ( x ι )) Explicit Characterization of I ( O ( x ι )) by Differential Radical Invariants Goal Automate the generation of Differential Radical Invariants K. Ghorbal (CMU) CMACS AVACS 22 CMACS 22 / 31

  24. Automated Generation Matrix Representation: Intuition invariant of degree 1 x 1 = a 1 x 1 + a 2 x 2 ˙ h = α 1 x 1 + α 2 x 2 + α 3 x 0 x 2 = b 1 x 1 + b 2 x 2 ˙ L p ( h ) = α 1 ( a 1 x 1 + a 2 x 2 ) + α 2 ( b 1 x 1 + b 2 x 2 ) L p ( h ) ∈ � h � if and only if ∃ β ∈ R s.t. L p ( h ) = β h     ( − a 1 + β ) α 1 + ( − b 1 ) α 2 = 0 − a 1 + β − b 1 0 α 1  = 0 ( − a 2 ) α 1 + ( − b 2 + β ) α 2 = 0 ↔ − a 2 − b 2 + β 0 α 2    ( β ) α 3 = 0 0 0 β α 3 K. Ghorbal (CMU) CMACS AVACS 23 CMACS 23 / 31

  25. Automated Generation Matrix Representation Explicit Ideal Membership N − 1 L ( N ) ( h ) ∈ � L (0) p ( h ) , . . . , L ( N − 1) ( h ) � ↔ L ( N ) g i L ( i ) � ( h ) = p ( h ) p p p i =0 � n + d � Polynomial ↔ Coefficients (up to monomial order) d h ↔ α = ( α 1 , α 2 , . . . , α r ) ↔ β i = ( β 1 , β 2 , . . . , β s i ) g i Matrix Representation N − 1 L ( N ) g i L ( i ) � ( h ) = p ( h ) ↔ M ( β ) α = 0 p i =0 α lies in the Kernel of M ( β ) def = { α ∈ R r | M ( β ) α = 0 } K. Ghorbal (CMU) CMACS AVACS 24 CMACS 24 / 31

Download Presentation
Download Policy: The content available on the website is offered to you 'AS IS' for your personal information and use only. It cannot be commercialized, licensed, or distributed on other websites without prior consent from the author. To download a presentation, simply click this link. If you encounter any difficulties during the download process, it's possible that the publisher has removed the file from their server.

Recommend


More recommend