Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott,...

95
Verified Computer-Aided Mathematics Assia Mahboubi Assia Mahboubi – Verified Computer-Aided Mathematics 1

Transcript of Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott,...

Page 1: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Verified Computer-AidedMathematics

Assia Mahboubi

Assia Mahboubi – Verified Computer-Aided Mathematics 1

Page 2: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

What is a proof assistant?

Mathematica: A System for Doing Mathematics by Computer,Stephen Wolfram, Addison-Wesley 1988.

Assia Mahboubi – Verified Computer-Aided Mathematics 2

Page 3: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

What is a proof assistant?

Mathematica:

A System for Doing Mathematics by Computer

,Stephen Wolfram, Addison-Wesley 1988.

Assia Mahboubi – Verified Computer-Aided Mathematics 2

Page 4: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

What is a proof assistant?

Mathematica: A System for Doing Mathematics by Computer,Stephen Wolfram, Addison-Wesley 1988.

Assia Mahboubi – Verified Computer-Aided Mathematics 2

Page 5: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Proof assistants

Representing mathematics in a fixed formal language:

• Define mathematical objects, statements, proofs;

• Verify the correctness of proofs by a routine process.

And use a proof assistant to make this possible in practice.

Assia Mahboubi – Verified Computer-Aided Mathematics 3

Page 6: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Proof assistants

Representing mathematics in a fixed formal language:

• Define mathematical objects, statements, proofs;

• Verify the correctness of proofs by a routine process.

And use a proof assistant to make this possible in practice.

Assia Mahboubi – Verified Computer-Aided Mathematics 3

Page 7: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Proof Assistants

Formal Logic

Proof Assistant

ProofChecker

Libraries

Assia Mahboubi – Verified Computer-Aided Mathematics 4

Page 8: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Assistance for building proofs

• Tactic language;

• Formal-proof producing decision procedures.

Assia Mahboubi – Verified Computer-Aided Mathematics 5

Page 9: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Assistance for writing statements

Variables gT (G : {group gT}).

Lemma cfCauchySchwarz (phi psi : ’CF(G)) :

|[phi, psi]| 2 <= [phi] * [psi] , eq_iff (~~ free (phi :: psi)).

Assia Mahboubi – Verified Computer-Aided Mathematics 6

Page 10: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Assistance for writing statements

Variables gT (G : {group gT}).

Lemma cfCauchySchwarz (phi psi : ’CF(G)) :

|[phi, psi]| 2 <= [phi] * [psi] , eq_iff (~~ free (phi :: psi)).

Assia Mahboubi – Verified Computer-Aided Mathematics 6

Page 11: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Usage

• Program verification:- Compcert compiler (Leroy et al., 2006),- sel4 micro-kernel (Klein et al., 2008),- etc.

• Mathematics:- Four color theorem (Gonthier, 2004),- Odd Order Theorem (Gonthier et al, 2012),- Flyspeck (Hales et al. 2014),- etc.

Assia Mahboubi – Verified Computer-Aided Mathematics 7

Page 12: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Usage

• Program verification:- Compcert compiler (Leroy et al., 2006),- sel4 micro-kernel (Klein et al., 2008),- etc.

• Mathematics:- Four color theorem (Gonthier, 2004),- Odd Order Theorem (Gonthier et al, 2012),- Flyspeck (Hales et al. 2014),- etc.

Assia Mahboubi – Verified Computer-Aided Mathematics 7

Page 13: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Assia Mahboubi – Verified Computer-Aided Mathematics 8

Page 14: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

In the Age of the Turing Machine

The Misfortunes of a Trio of Mathematicians Using ComputerAlgebra Systems. Can We Trust in Them?

Antonio J. Duran, Mario Perez, and Juan L. Varona, Notices of the AMS, Nov. 2014

Assia Mahboubi – Verified Computer-Aided Mathematics 9

Page 15: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

In the Age of the Turing Machine

The Misfortunes of a Trio of Mathematicians Using ComputerAlgebra Systems. Can We Trust in Them?

Antonio J. Duran, Mario Perez, and Juan L. Varona, Notices of the AMS, Nov. 2014

Assia Mahboubi – Verified Computer-Aided Mathematics 9

Page 16: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Computing Integrals

m ≤∫ b

a

f ≤ M

∫ 8

0

sin(t + exp(t))dt,

∫ 2

√π

ln(t)dt,

∫ +∞

0

cos(t)ln(t)

t2dt

Joint work with Guillaume Melquiond and Thomas Sibut-Pinote, JAR 2018

Assia Mahboubi – Verified Computer-Aided Mathematics 10

Page 17: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Computing Integrals

m ≤∫ b

a

f ≤ M

∫ 8

0

sin(t + exp(t))dt,

∫ 2

√π

ln(t)dt,

∫ +∞

0

cos(t)ln(t)

t2dt

Joint work with Guillaume Melquiond and Thomas Sibut-Pinote, JAR 2018

Assia Mahboubi – Verified Computer-Aided Mathematics 10

Page 18: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Example

Assia Mahboubi – Verified Computer-Aided Mathematics 11

Page 19: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Ternary Goldbach Conjecture

Assia Mahboubi – Verified Computer-Aided Mathematics 12

Page 20: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Integral computations

Assia Mahboubi – Verified Computer-Aided Mathematics 13

Page 21: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Real numbers in a computer: Floats

• Compromise between precision and range• IEEE standards• Efficient algorithms

• Compromise between precision and range• Notoriously difficult to specify

Credit: Dominique ThiebautAssia Mahboubi – Verified Computer-Aided Mathematics 14

Page 22: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Real numbers in a computer: intervals

I = {[a, b] | a, b ∈ R ∪ {+∞} ∪ {−∞}}

• Examples:

[0, 0] [3.14, 3.15] [17,+∞] [−∞,+∞]

• Operations:

[−1, 2]× [−3, 1] = [−6, 2] [0, 100]− [0, 100] = [−100, 100]

• Computations: floating-point numbers as bounds.

• Reasoning: interval analysis.

Assia Mahboubi – Verified Computer-Aided Mathematics 15

Page 23: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Interval extensions

F : In → I is an interval extension of f : Rn → R, if:

∀x ∈ Rn,∀x ∈ In, x ∈ x⇒ f (x) ∈ F(x)

An interval function F : In → I is isotonic if:

yi ⊂ xi ⇒ F(y1, . . . , yn) ⊂ F(x1, . . . , xn)

[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud]

[Validated Numerics: A Short Introduction to Rigorous Computations - W. Tucker]

Assia Mahboubi – Verified Computer-Aided Mathematics 16

Page 24: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Interval extensions

F : In → I is an interval extension of f : Rn → R, if:

∀x ∈ Rn,∀x ∈ In, x ∈ x⇒ f (x) ∈ F(x)

An interval function F : In → I is isotonic if:

yi ⊂ xi ⇒ F(y1, . . . , yn) ⊂ F(x1, . . . , xn)

[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud]

[Validated Numerics: A Short Introduction to Rigorous Computations - W. Tucker]

Assia Mahboubi – Verified Computer-Aided Mathematics 16

Page 25: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Interval Riemann integration

If f is continuous on [a, b], then:∫ b

a

f ∈ (b − a)[m,M]

If F extends f , and a ∈ a, b ∈ b:∫ b

a

f ∈ (b− a)F(hull(a,b))

Combine with dichotomy and higher order methods.

Assia Mahboubi – Verified Computer-Aided Mathematics 17

Page 26: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Interval Riemann integration

If f is continuous on [a, b], then:∫ b

a

f ∈ (b − a)[m,M]

If F extends f , and a ∈ a, b ∈ b:

∫ b

a

f ∈ (b− a)F(hull(a,b))

Combine with dichotomy and higher order methods.

Assia Mahboubi – Verified Computer-Aided Mathematics 17

Page 27: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Interval Riemann integration

If f is continuous on [a, b], then:∫ b

a

f ∈ (b − a)[m,M]

If F extends f , and a ∈ a, b ∈ b:∫ b

a

f ∈ (b− a)F(hull(a,b))

Combine with dichotomy and higher order methods.

Assia Mahboubi – Verified Computer-Aided Mathematics 17

Page 28: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Interval Riemann integration

If f is continuous on [a, b], then:∫ b

a

f ∈ (b − a)[m,M]

If F extends f , and a ∈ a, b ∈ b:∫ b

a

f ∈ (b− a)F(hull(a,b))

Combine with dichotomy and higher order methods.

Assia Mahboubi – Verified Computer-Aided Mathematics 17

Page 29: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Correctness theorem

• Program: cos : I→ I

• Mathematical definition: cos : R→ R• Specification:

For any x ∈ R and any i ∈ I, x ∈ i⇒ cos(x) ∈ cos(i)

⇒ A formal proof per integrand.

Assia Mahboubi – Verified Computer-Aided Mathematics 18

Page 30: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Correctness theorem

• Program: cos : I→ I• Mathematical definition: cos : R→ R

• Specification:

For any x ∈ R and any i ∈ I, x ∈ i⇒ cos(x) ∈ cos(i)

⇒ A formal proof per integrand.

Assia Mahboubi – Verified Computer-Aided Mathematics 18

Page 31: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Correctness theorem

• Program: cos : I→ I• Mathematical definition: cos : R→ R• Specification:

For any x ∈ R and any i ∈ I, x ∈ i⇒ cos(x) ∈ cos(i)

⇒ A formal proof per integrand.

Assia Mahboubi – Verified Computer-Aided Mathematics 18

Page 32: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Correctness theorem

• Program: cos : I→ I• Mathematical definition: cos : R→ R• Specification:

For any x ∈ R and any i ∈ I, x ∈ i⇒ cos(x) ∈ cos(i)

⇒ A formal proof per integrand.

Assia Mahboubi – Verified Computer-Aided Mathematics 18

Page 33: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Catalogue

Based on an abstract syntax of (univariate) expressions:

E := x | F| π |E + E | E − E | E × E | E ÷ E | − E | ‖E‖ |√E | Ek |

cos(E) | sin(E) | tan(E) | atan(E) |exp(E) | ln(E)

And on an arithmetic of interval extensions.[Proving bounds on real-valued functions with computations - G. Melquiond, Proc. of IJCAR 2008]

Assia Mahboubi – Verified Computer-Aided Mathematics 19

Page 34: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Computations of extensions

Expressions are equipped with several evaluations schemes:

• [e]R : R→ R• [e]R⊥ : R⊥ → R⊥• [e]I⊥ : I⊥ → I⊥

• Value ⊥R models the complement of the definition domain.

• Interval ⊥I is the only one containing ⊥R.

• A correctness theorem ensures:

∀e ∈ E ,∀i ∈ I⊥,∀x ∈ i, [e]R⊥(x) ∈ [e]I⊥(i)

Assia Mahboubi – Verified Computer-Aided Mathematics 20

Page 35: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Computations of extensions

Expressions are equipped with several evaluations schemes:

• [e]R : R→ R• [e]R⊥ : R⊥ → R⊥• [e]I⊥ : I⊥ → I⊥

• Value ⊥R models the complement of the definition domain.

• Interval ⊥I is the only one containing ⊥R.

• A correctness theorem ensures:

∀e ∈ E ,∀i ∈ I⊥,∀x ∈ i, [e]R⊥(x) ∈ [e]I⊥(i)

Assia Mahboubi – Verified Computer-Aided Mathematics 20

Page 36: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Computations of extensions

Expressions are equipped with several evaluations schemes:

• [e]R : R→ R• [e]R⊥ : R⊥ → R⊥• [e]I⊥ : I⊥ → I⊥

• Value ⊥R models the complement of the definition domain.

• Interval ⊥I is the only one containing ⊥R.

• A correctness theorem ensures:

∀e ∈ E ,∀i ∈ I⊥,∀x ∈ i, [e]R⊥(x) ∈ [e]I⊥(i)

Assia Mahboubi – Verified Computer-Aided Mathematics 20

Page 37: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Computations of extensions

Expressions are equipped with several evaluations schemes:

• [e]R : R→ R• [e]R⊥ : R⊥ → R⊥• [e]I⊥ : I⊥ → I⊥

• Value ⊥R models the complement of the definition domain.

• Interval ⊥I is the only one containing ⊥R.

• A correctness theorem ensures:

∀e ∈ E ,∀i ∈ I⊥,∀x ∈ i, [e]R⊥(x) ∈ [e]I⊥(i)

Assia Mahboubi – Verified Computer-Aided Mathematics 20

Page 38: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Meaning

∫ b

a

f ?

In fact, for any expression e ∈ E and interval i:

[e]I⊥(i) 6= ⊥I ⇒ f is continuous on i

⇒ If∫ b

af can be approximated, then f is integrable on [a, b].

Assia Mahboubi – Verified Computer-Aided Mathematics 21

Page 39: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Meaning

∫ b

a

[e]R?

In fact, for any expression e ∈ E and interval i:

[e]I⊥(i) 6= ⊥I ⇒ f is continuous on i

⇒ If∫ b

af can be approximated, then f is integrable on [a, b].

Assia Mahboubi – Verified Computer-Aided Mathematics 21

Page 40: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Meaning

∫ b

a

[e]R?

In fact, for any expression e ∈ E and interval i:

[e]I⊥(i) 6= ⊥I ⇒ f is continuous on i

⇒ If∫ b

af can be approximated, then f is integrable on [a, b].

Assia Mahboubi – Verified Computer-Aided Mathematics 21

Page 41: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Benchmarks

∫ 1

0

∣∣(x4 + 10x3 + 19x2 − 6x − 6)ex∣∣ dx ' 11.14731055005714

Error Time Accuracy Degree Depth Precision10−3 0.7 14 5 8 3010−6 0.9 24 6 13 4010−9 1.3 34 8 18 5010−12 1.9 44 10 22 6010−15 2.7 54 12 28 70

Assia Mahboubi – Verified Computer-Aided Mathematics 22

Page 42: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Benchmarks

∫ 1

0

∣∣(x4 + 10x3 + 19x2 − 6x − 6)ex∣∣ dx ' 11.14731055005714

• Octave’s quad/quadgk: only 10/9 correct digits;

• INTLAB verifyquad: false answer without warning;

• VNODE-LP: cannot be used because of the absolute value.

Assia Mahboubi – Verified Computer-Aided Mathematics 23

Page 43: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Benchmarks

∫ +∞

100000

1 +(

0.5·ln(1+2.25/τ2)+4.1396+lnπln τ

)2

1 + 0.25/τ2·

ln2 τ

τ2dτ ' −3.2555895745 · 10−6

Error Time Accuracy Degree Depth Precision

10−3 0.6 2 3 0 3010−4 0.8 5 5 2 3010−5 1.5 8 7 6 3010−6 3.1 11 9 11 3010−7 5.6 15 12 12 3010−8 11.2 18 15 15 30

Thus: ∫ +∞

−∞

1 +(

0.5·ln(1+2.25/τ2)+4.1396+lnπln τ

)2

1 + 0.25/τ2·

ln2 τ

τ2dτ ∈ [226.849; 226.850]

The upper bound 226.844 given in the preprint is incorrect.

Assia Mahboubi – Verified Computer-Aided Mathematics 24

Page 44: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Benchmarks

∫ +∞

100000

1 +(

0.5·ln(1+2.25/τ2)+4.1396+lnπln τ

)2

1 + 0.25/τ2·

ln2 τ

τ2dτ ' −3.2555895745 · 10−6

Error Time Accuracy Degree Depth Precision

10−3 0.6 2 3 0 3010−4 0.8 5 5 2 3010−5 1.5 8 7 6 3010−6 3.1 11 9 11 3010−7 5.6 15 12 12 3010−8 11.2 18 15 15 30

Thus: ∫ +∞

−∞

1 +(

0.5·ln(1+2.25/τ2)+4.1396+lnπln τ

)2

1 + 0.25/τ2·

ln2 τ

τ2dτ ∈ [226.849; 226.850]

The upper bound 226.844 given in the preprint is incorrect.

Assia Mahboubi – Verified Computer-Aided Mathematics 24

Page 45: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

More

• Formal Verification of Nonlinear Inequalities with TaylorInterval Approximations

Thomas Hales and Alexey Solovyev, Proc. of NFM 2013.

• A Verified ODE Solver and the Lorenz AttractorFabian Immler, JAR 2018.

Assia Mahboubi – Verified Computer-Aided Mathematics 25

Page 46: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

More

• Formal Verification of Nonlinear Inequalities with TaylorInterval Approximations

Thomas Hales and Alexey Solovyev, Proc. of NFM 2013.

• A Verified ODE Solver and the Lorenz AttractorFabian Immler, JAR 2018.

Assia Mahboubi – Verified Computer-Aided Mathematics 25

Page 47: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Special functions and sequences

Abramowitz and Stegun Plouffe and Sloane

Assia Mahboubi – Verified Computer-Aided Mathematics 26

Page 48: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

∂-finiteness

• A function f : K→ K is D-finite if for all x :

pr (x)f (r)(x) + · · ·+ p1(x)f ′(x) + p0(x)f (x) = 0

• A sequence u ∈ KN is P-recursive if for all n ∈ N

pr (n)un+r + · · ·+ p0(n)un = 0

where pi ∈ C[n] are polynomial coefficients and pr 6= 0.

⇒ Extends to systems and multivariates sequences/functions.

Assia Mahboubi – Verified Computer-Aided Mathematics 27

Page 49: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

∂-finiteness

• A function f : K→ K is D-finite if for all x :

pr (x)f (r)(x) + · · ·+ p1(x)f ′(x) + p0(x)f (x) = 0

• A sequence u ∈ KN is P-recursive if for all n ∈ N

pr (n)un+r + · · ·+ p0(n)un = 0

where pi ∈ C[n] are polynomial coefficients and pr 6= 0.

⇒ Extends to systems and multivariates sequences/functions.

Assia Mahboubi – Verified Computer-Aided Mathematics 27

Page 50: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

∂-finiteness

• A function f : K→ K is D-finite if for all x :

pr (x)f (r)(x) + · · ·+ p1(x)f ′(x) + p0(x)f (x) = 0

• A sequence u ∈ KN is P-recursive if for all n ∈ N

pr (n)un+r + · · ·+ p0(n)un = 0

where pi ∈ C[n] are polynomial coefficients and pr 6= 0.

⇒ Extends to systems and multivariates sequences/functions.

Assia Mahboubi – Verified Computer-Aided Mathematics 27

Page 51: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Examples

• Bessel functions:

x2d2f

dx+ x

df

dx+ (x2 − a2)f = 0

• Binomial coefficients(nm

):

un+1,m+1, = un+1,m + un,m+1 with u0,0 = un,n = 1

Assia Mahboubi – Verified Computer-Aided Mathematics 28

Page 52: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Closure properties

P-recursive sequences on a field K form a K-algebra.

If u and v are P-recursive it is possible to:

• compute recurrences canceling (u + v)

• compute recurrences canceling (u ∗ v)

For (un,k) P-recursive, it is possible to:

• Compute a recurrence canceling Un :=∑n

k=0 un,k

Assia Mahboubi – Verified Computer-Aided Mathematics 29

Page 53: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Closure properties

P-recursive sequences on a field K form a K-algebra.

If u and v are P-recursive it is possible to:

• compute recurrences canceling (u + v)

• compute recurrences canceling (u ∗ v)

For (un,k) P-recursive, it is possible to:

• Compute a recurrence canceling Un :=∑n

k=0 un,k

Assia Mahboubi – Verified Computer-Aided Mathematics 29

Page 54: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Irrationality of ζ(3)

Let ζ(3) be the real number∑+∞

k=11k3 .

Theorem (Apery, 1978): The constant ζ(3) is irrational.

See:“A proof that Euler missed: Apery’s proof of the irrationality ofζ(3). An informal report.” by A. van der Poorten, 1979.

for the tale of this proof, and of its verification.

Assia Mahboubi – Verified Computer-Aided Mathematics 30

Page 55: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Irrationality of ζ(3)

The crux in Apery’s proof is to verify that (an)n∈N and (bn)n∈Nverify the recurrence:

n3yn − (34n3 − 51n2 + 27n − 5)yn−1 + (n − 1)3yn−2 = 0

with:

an =n∑

k=0

(nk

)2(n+kk

)2, bn = an

n∑k=1

1

k3+

n∑k=1

k∑m=1

(−1)m+1(nk

)2(n+kk

)2

2m3(nm

)(n+mm

) .

Assia Mahboubi – Verified Computer-Aided Mathematics 31

Page 56: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Irrationality of ζ(3)

The crux in Apery’s proof is to verify that (an)n∈N and (bn)n∈Nverify the recurrence:

n3yn − (34n3 − 51n2 + 27n − 5)yn−1 + (n − 1)3yn−2 = 0

with:

an =n∑

k=0

(nk

)2(n+kk

)2, bn = an

n∑k=1

1

k3+

n∑k=1

k∑m=1

(−1)m+1(nk

)2(n+kk

)2

2m3(nm

)(n+mm

) .

Assia Mahboubi – Verified Computer-Aided Mathematics 31

Page 57: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Checking the recurrence: telescopes

To prove that an :=∑

k vn,k with vn,k :=(nk

)2(n+kk

)2verifies:

n3yn − (34n3 − 51n2 + 27n − 5)yn−1 + (n − 1)3yn−2 = 0

Cohen and Zagier constructed wn,k such that:

n3vn,k−(34n3−51n2+27n−5)vn−1,k+(n−1)3vn−2,k = wn,k+1 − wn,k

And then they sum over k both hand sides:∑k

(n3vn,k−(34n3−51n2+27n−5)vn−1,k+(n−1)3vn−2,k) =∑k

(wn,k+1 − wn,k)

n3an − (34n3 − 51n2 + 27n − 5)an−1 + (n − 1)3an−2 = wn,∞ − wn,0 = 0

Assia Mahboubi – Verified Computer-Aided Mathematics 32

Page 58: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Checking the recurrence: telescopes

To prove that an :=∑

k vn,k with vn,k :=(nk

)2(n+kk

)2verifies:

n3yn − (34n3 − 51n2 + 27n − 5)yn−1 + (n − 1)3yn−2 = 0

Cohen and Zagier constructed wn,k such that:

n3vn,k−(34n3−51n2+27n−5)vn−1,k+(n−1)3vn−2,k = wn,k+1 − wn,k

And then they sum over k both hand sides:∑k

(n3vn,k−(34n3−51n2+27n−5)vn−1,k+(n−1)3vn−2,k) =∑k

(wn,k+1 − wn,k)

n3an − (34n3 − 51n2 + 27n − 5)an−1 + (n − 1)3an−2 = wn,∞ − wn,0 = 0

Assia Mahboubi – Verified Computer-Aided Mathematics 32

Page 59: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Checking the recurrence: telescopes

To prove that an :=∑

k vn,k with vn,k :=(nk

)2(n+kk

)2verifies:

n3yn − (34n3 − 51n2 + 27n − 5)yn−1 + (n − 1)3yn−2 = 0

Cohen and Zagier constructed wn,k such that:

n3vn,k−(34n3−51n2+27n−5)vn−1,k+(n−1)3vn−2,k = wn,k+1 − wn,k

And then they sum over k both hand sides:

∑k

(n3vn,k−(34n3−51n2+27n−5)vn−1,k+(n−1)3vn−2,k) =∑k

(wn,k+1 − wn,k)

n3an − (34n3 − 51n2 + 27n − 5)an−1 + (n − 1)3an−2 = wn,∞ − wn,0 = 0

Assia Mahboubi – Verified Computer-Aided Mathematics 32

Page 60: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Checking the recurrence: telescopes

To prove that an :=∑

k vn,k with vn,k :=(nk

)2(n+kk

)2verifies:

n3yn − (34n3 − 51n2 + 27n − 5)yn−1 + (n − 1)3yn−2 = 0

Cohen and Zagier constructed wn,k such that:

n3vn,k−(34n3−51n2+27n−5)vn−1,k+(n−1)3vn−2,k = wn,k+1 − wn,k

And then they sum over k both hand sides:∑k

(n3vn,k−(34n3−51n2+27n−5)vn−1,k+(n−1)3vn−2,k) =∑k

(wn,k+1 − wn,k)

n3an − (34n3 − 51n2 + 27n − 5)an−1 + (n − 1)3an−2 = wn,∞ − wn,0 = 0

Assia Mahboubi – Verified Computer-Aided Mathematics 32

Page 61: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Checking the recurrence: telescopes

To prove that an :=∑

k vn,k with vn,k :=(nk

)2(n+kk

)2verifies:

n3yn − (34n3 − 51n2 + 27n − 5)yn−1 + (n − 1)3yn−2 = 0

Cohen and Zagier constructed wn,k such that:

n3vn,k−(34n3−51n2+27n−5)vn−1,k+(n−1)3vn−2,k = wn,k+1 − wn,k

And then they sum over k both hand sides:∑k

(n3vn,k−(34n3−51n2+27n−5)vn−1,k+(n−1)3vn−2,k) =∑k

(wn,k+1 − wn,k)

n3an − (34n3 − 51n2 + 27n − 5)an−1 + (n − 1)3an−2

=

wn,∞ − wn,0 = 0

Assia Mahboubi – Verified Computer-Aided Mathematics 32

Page 62: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Checking the recurrence: telescopes

To prove that an :=∑

k vn,k with vn,k :=(nk

)2(n+kk

)2verifies:

n3yn − (34n3 − 51n2 + 27n − 5)yn−1 + (n − 1)3yn−2 = 0

Cohen and Zagier constructed wn,k such that:

n3vn,k−(34n3−51n2+27n−5)vn−1,k+(n−1)3vn−2,k = wn,k+1 − wn,k

And then they sum over k both hand sides:∑k

(n3vn,k−(34n3−51n2+27n−5)vn−1,k+(n−1)3vn−2,k) =∑k

(wn,k+1 − wn,k)

n3an − (34n3 − 51n2 + 27n − 5)an−1 + (n − 1)3an−2 =

wn,∞ − wn,0 = 0

Assia Mahboubi – Verified Computer-Aided Mathematics 32

Page 63: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Checking the recurrence: telescopes

To prove that an :=∑

k vn,k with vn,k :=(nk

)2(n+kk

)2verifies:

n3yn − (34n3 − 51n2 + 27n − 5)yn−1 + (n − 1)3yn−2 = 0

Cohen and Zagier constructed wn,k such that:

n3vn,k−(34n3−51n2+27n−5)vn−1,k+(n−1)3vn−2,k = wn,k+1 − wn,k

And then they sum over k both hand sides:∑k

(n3vn,k−(34n3−51n2+27n−5)vn−1,k+(n−1)3vn−2,k) =∑k

(wn,k+1 − wn,k)

n3an − (34n3 − 51n2 + 27n − 5)an−1 + (n − 1)3an−2 = wn,∞ − wn,0 = 0

Assia Mahboubi – Verified Computer-Aided Mathematics 32

Page 64: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Alternate proofs

• Paper proofs:Beukers (1979, 1987), Nesterenko (1996), Rajkumar(2012), . . .

• Algorithmic proofs:Zeilberger (1993), Zudilin (2002, 2009), Salvy (2003),Schneider (2007), . . .

Assia Mahboubi – Verified Computer-Aided Mathematics 33

Page 65: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Formal recurrence operators

We denote A = Q(n, k)〈Sn, Sk〉 the ring of skew polynomials:

• in the (commuting) indeterminates Sn and Sk

• with coefficients in Q(n, k)

• with the commutation rule:

S inS

jkc(n, k) = c(n + i , k + j)S i

nSjk .

Elements of A can be viewed as algebraic representations forlinear recurrence operators with rational function coefficients.

Assia Mahboubi – Verified Computer-Aided Mathematics 34

Page 66: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Formal recurrence operators

Consider

• P =∑

(i ,j)∈I pi ,j(n, k)S inS

jk ∈ A,

• f a sequence.

The operator P acts on f by:

(P · f )n,k =∑

(i ,j)∈I

pi ,j(n, k)fn+i ,k+j

Assia Mahboubi – Verified Computer-Aided Mathematics 35

Page 67: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Example: binomial coefficient

Recurrences for the binomial coefficient(nk

)(

n + 1

k

)− n + 1

n + 1− k

(n

k

)= 0,

(n

k + 1

)− n − k

k + 1

(n

k

)= 0

can be represented using the operators:

P1 = Sn −n + 1

n + 1− k1, P2 = Sk −

n − k

k + 11

as:P1 · f = 0, P2 · f = 0

Assia Mahboubi – Verified Computer-Aided Mathematics 36

Page 68: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Example: binomial coefficient

Recurrences for the binomial coefficient(nk

)(

n + 1

k

)− n + 1

n + 1− k

(n

k

)= 0,

(n

k + 1

)− n − k

k + 1

(n

k

)= 0

can be represented using the operators:

P1 = Sn −n + 1

n + 1− k, P2 = Sk −

n − k

k + 1as:

P1 · f = 0, P2 · f = 0

Assia Mahboubi – Verified Computer-Aided Mathematics 37

Page 69: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Creative telescoping

The aim is:

• starting from sequence (fn,k),

• to compute annihilating operators for Fn =∑n

k=0 fn,k .

For this:

• An algorithm finds a pair (P ,Q) satisfying

P · f = (Sk − 1)Q · f ;

• With Q ∈ A and P ∈ Q(n, k)〈Sn〉.

Assia Mahboubi – Verified Computer-Aided Mathematics 38

Page 70: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Creative telescoping

The aim is:

• starting from sequence (fn,k),

• to compute annihilating operators for Fn =∑n

k=0 fn,k .

For this:

• An algorithm finds a pair (P ,Q) satisfying

P · f = (Sk − 1)Q · f ;

• With Q ∈ A and P ∈ Q(n, k)〈Sn〉.

Assia Mahboubi – Verified Computer-Aided Mathematics 38

Page 71: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Annihilating ideal and Grobner basis

Let f a (germ of) sequence, cancelled by a certain Pf ∈ A.

• {P ∈ A : P · f = 0} is a left ideal of A;

• Algorithms exist to compute Grobner bases of such an ideal.

• Such a Grobner basis is a good representation for f itself.

Assia Mahboubi – Verified Computer-Aided Mathematics 39

Page 72: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Checking a candidate recurrence

Grobner bases are excellent tools to check the validity of acandidate recurrence.

Check: (n + 1

k + 1

)−(

n

k + 1

)−(n

k

)= 0

Using:(n + 1

k

)=

n + 1

n + 1− k

(n

k

),

(n

k + 1

)=

n − k

k + 1

(n

k

)

Assia Mahboubi – Verified Computer-Aided Mathematics 40

Page 73: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Computing Apery’s recurrence

an =n∑

k=0

(nk

)2(n+kk

)2, bn =

n∑k=1

(nk

)2(n+kk

)2

n∑m=1

1

m3+

k∑m=1

(−1)m+1

2m3(nm

)(n+mm

) .

step explicit form system operation input(s)

1 cn,k =(nk

)2(n+kk

)2C direct

2 an =∑n

k=1 cn,k A Σ C

3 dn,m = (−1)m+1

2m3(nm)(n+m

m )D direct

4 sn,k =∑k

m=1 dn,m S Σ D

5 zn =∑n

m=11m3 Z direct

6 un,k = zn + sn,k U + Z and S

7 vn,k = cn,kun,k V × C and U

8 bn =∑n

k=1 vn,k B Σ V

Assia Mahboubi – Verified Computer-Aided Mathematics 41

Page 74: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Computing Apery’s recurrence

an =n∑

k=0

(nk

)2(n+kk

)2, bn =

n∑k=1

(nk

)2(n+kk

)2

n∑m=1

1

m3+

k∑m=1

(−1)m+1

2m3(nm

)(n+mm

) .

step explicit form system operation input(s)

1 cn,k =(nk

)2(n+kk

)2C direct

2 an =∑n

k=1 cn,k A Σ C

3 dn,m = (−1)m+1

2m3(nm)(n+m

m )D direct

4 sn,k =∑k

m=1 dn,m S Σ D

5 zn =∑n

m=11m3 Z direct

6 un,k = zn + sn,k U + Z and S

7 vn,k = cn,kun,k V × C and U

8 bn =∑n

k=1 vn,k B Σ V

Assia Mahboubi – Verified Computer-Aided Mathematics 42

Page 75: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Checking Apery’s recurrence

an =n∑

k=0

(nk

)2(n+kk

)2, bn =

n∑k=1

(nk

)2(n+kk

)2

n∑m=1

1

m3+

k∑m=1

(−1)m+1

2m3(nm

)(n+mm

) .

step explicit form GB operation input(s)8 bn =

∑nk=1 vn,k B creative telescoping V

7 vn,k = cn,kun,k V product C and U

6 un,k = zn + sn,k U addition Z and S

5 zn

=∑n

m=11m3

Z

direct

4 sn,k =∑k

m=1 dn,m S creative telescoping D

3 dn,m

= (−1)m+1

2m3(nm)(n+m

m )

D

direct

2 an =∑n

k=1 cn,k A creative telescoping C

1 cn,k

=(nk

)2(n+kk

)2

C

direct

Assia Mahboubi – Verified Computer-Aided Mathematics 43

Page 76: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Checking Apery’s recurrence

an =n∑

k=0

(nk

)2(n+kk

)2, bn =

n∑k=1

(nk

)2(n+kk

)2

n∑m=1

1

m3+

k∑m=1

(−1)m+1

2m3(nm

)(n+mm

) .

step explicit form GB operation input(s)8 bn =

∑nk=1 vn,k B creative telescoping V

7 vn,k = cn,kun,k V product C and U

6 un,k = zn + sn,k U addition Z and S

5 zn =∑n

m=11m3 Z direct

4 sn,k =∑k

m=1 dn,m S creative telescoping D

3 dn,m = (−1)m+1

2m3(nm)(n+m

m )D direct

2 an =∑n

k=1 cn,k A creative telescoping C

1 cn,k =(nk

)2(n+kk

)2C direct

Assia Mahboubi – Verified Computer-Aided Mathematics 43

Page 77: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Sound creative telescoping

From the guarded creative telescoping identity for (fn,k)

(n, k) /∈ ∆⇒ (P · f ,k)n = (Q · f )n,k+1 − (Q · f )n,k ,

We can prove for Fn =∑n+β

k=α fn,k that:

(P · F )n =(

(Q · f )n,n+β+1 − (Q · f )n,α)

+r∑

i=1

i∑j=1

pi(n) fn+i ,n+β+j

+∑

α≤k≤n+β ∧ (n,k)∈∆

(P · f ,k)n − (Q · f )n,k+1 + (Q · f )n,k ,

Assia Mahboubi – Verified Computer-Aided Mathematics 44

Page 78: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Sound creative telescoping

From the guarded creative telescoping identity for (fn,k)

(n, k) /∈ ∆⇒ (P · f ,k)n = (Q · f )n,k+1 − (Q · f )n,k ,

We can prove for Fn =∑n+β

k=α fn,k that:

(P · F )n =(

(Q · f )n,n+β+1 − (Q · f )n,α)

+r∑

i=1

i∑j=1

pi(n) fn+i ,n+β+j

+∑

α≤k≤n+β ∧ (n,k)∈∆

(P · f ,k)n − (Q · f )n,k+1 + (Q · f )n,k ,

Assia Mahboubi – Verified Computer-Aided Mathematics 44

Page 79: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

A formal, computer-algebra aided, proof

Joint work with Frederic Chyzak, Thomas Sibut-Pinote, ITP 2014

• A Maple session computes the recurrence operators;

• It generates Coq statements of candidate properties;

• We annotate the generated recurrences with provisos;

• We prove lemmas by interactive formal proofs.

See also: Formal proofs of hypergeometric sums, John Harrison, Journal of Automated Reasoning, 2015

Assia Mahboubi – Verified Computer-Aided Mathematics 45

Page 80: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

A formal, computer-algebra aided, proof

Joint work with Frederic Chyzak, Thomas Sibut-Pinote, ITP 2014

• A Maple session computes the recurrence operators;

• It generates Coq statements of candidate properties;

• We annotate the generated recurrences with provisos;

• We prove lemmas by interactive formal proofs.

See also: Formal proofs of hypergeometric sums, John Harrison, Journal of Automated Reasoning, 2015

Assia Mahboubi – Verified Computer-Aided Mathematics 45

Page 81: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

A formal, computer-algebra aided, proof

Joint work with Frederic Chyzak, Thomas Sibut-Pinote, ITP 2014

• A Maple session computes the recurrence operators;

• It generates Coq statements of candidate properties;

• We annotate the generated recurrences with provisos;

• We prove lemmas by interactive formal proofs.

See also: Formal proofs of hypergeometric sums, John Harrison, Journal of Automated Reasoning, 2015

Assia Mahboubi – Verified Computer-Aided Mathematics 45

Page 82: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

First order theories

Inductive term : Type :=

| Var (n : N)| Add (t1 t2 : term)

| Opp (t : term)

| NatConst (n : N)| NatMul (t : term)(n : N)| Mul (t1 t2 : term)

.

Inductive formula : Type :=

| Bool (b : B)| Equal (t1 t2 : term)

| Leq (t1 t2 : term)

| And (f1 f2 : formula)

| Or (f1 f2 : formula)

| Implies (f1 f2 : formula)

| Not (f : formula)

| Exists (n : N) (f : formula)

| Forall (n : N) (f : formula)

.

Definition qfree_form : formula -> B

Definition eval (R : ringType) : seq R -> term -> R

Definition holds (R : ringType) : seq R -> formula -> Prop

⇒ (holds R e f) is the (yet unproved) statement “R |=e f ”.

Assia Mahboubi – Verified Computer-Aided Mathematics 46

Page 83: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

First order theories

Inductive term : Type :=

| Var (n : N)| Add (t1 t2 : term)

| Opp (t : term)

| NatConst (n : N)| NatMul (t : term)(n : N)| Mul (t1 t2 : term)

.

Inductive formula : Type :=

| Bool (b : B)| Equal (t1 t2 : term)

| Leq (t1 t2 : term)

| And (f1 f2 : formula)

| Or (f1 f2 : formula)

| Implies (f1 f2 : formula)

| Not (f : formula)

| Exists (n : N) (f : formula)

| Forall (n : N) (f : formula)

.

Definition qfree_form : formula -> B

Definition eval (R : ringType) : seq R -> term -> R

Definition holds (R : ringType) : seq R -> formula -> Prop

⇒ (holds R e f) is the (yet unproved) statement “R |=e f ”.

Assia Mahboubi – Verified Computer-Aided Mathematics 46

Page 84: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

First order theories

Inductive term : Type :=

| Var (n : N)| Add (t1 t2 : term)

| Opp (t : term)

| NatConst (n : N)| NatMul (t : term)(n : N)| Mul (t1 t2 : term)

.

Inductive formula : Type :=

| Bool (b : B)| Equal (t1 t2 : term)

| Leq (t1 t2 : term)

| And (f1 f2 : formula)

| Or (f1 f2 : formula)

| Implies (f1 f2 : formula)

| Not (f : formula)

| Exists (n : N) (f : formula)

| Forall (n : N) (f : formula)

.

Definition qfree_form : formula -> B

Definition eval (R : ringType) : seq R -> term -> R

Definition holds (R : ringType) : seq R -> formula -> Prop

⇒ (holds R e f) is the (yet unproved) statement “R |=e f ”.

Assia Mahboubi – Verified Computer-Aided Mathematics 46

Page 85: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

First order theories

Inductive term : Type :=

| Var (n : N)| Add (t1 t2 : term)

| Opp (t : term)

| NatConst (n : N)| NatMul (t : term)(n : N)| Mul (t1 t2 : term)

.

Inductive formula : Type :=

| Bool (b : B)| Equal (t1 t2 : term)

| Leq (t1 t2 : term)

| And (f1 f2 : formula)

| Or (f1 f2 : formula)

| Implies (f1 f2 : formula)

| Not (f : formula)

| Exists (n : N) (f : formula)

| Forall (n : N) (f : formula)

.

Definition qfree_form : formula -> B

Definition eval (R : ringType) : seq R -> term -> R

Definition holds (R : ringType) : seq R -> formula -> Prop

⇒ (holds R e f) is the (yet unproved) statement “R |=e f ”.

Assia Mahboubi – Verified Computer-Aided Mathematics 46

Page 86: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

First order theories

Inductive term : Type :=

| Var (n : N)| Add (t1 t2 : term)

| Opp (t : term)

| NatConst (n : N)| NatMul (t : term)(n : N)| Mul (t1 t2 : term)

.

Inductive formula : Type :=

| Bool (b : B)| Equal (t1 t2 : term)

| Leq (t1 t2 : term)

| And (f1 f2 : formula)

| Or (f1 f2 : formula)

| Implies (f1 f2 : formula)

| Not (f : formula)

| Exists (n : N) (f : formula)

| Forall (n : N) (f : formula)

.

Definition qfree_form : formula -> B

Definition eval (R : ringType) : seq R -> term -> R

Definition holds (R : ringType) : seq R -> formula -> Prop

⇒ (holds R e f) is the (yet unproved) statement “R |=e f ”.

Assia Mahboubi – Verified Computer-Aided Mathematics 46

Page 87: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Quantifier elimination: real closed fields

⇒ A function:

Definition rcf_qelim : formula -> formula

such that:

• forall f : formula, qfree (qelim f)= true

• forall (R : realClosedField) (e : seq R) (f : formula),

holds e f <-> holds e (rcf_qelim e f)

Joint work with Cyril Cohen, LMCS 2012.

⇒ A (constructive) theory of semi-algebraic sets and functions.Boris Djalal, Proceedings of CPP 2018.

Assia Mahboubi – Verified Computer-Aided Mathematics 47

Page 88: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Quantifier elimination: real closed fields

⇒ A function:

Definition rcf_qelim : formula -> formula

such that:

• forall f : formula, qfree (qelim f)= true

• forall (R : realClosedField) (e : seq R) (f : formula),

holds e f <-> holds e (rcf_qelim e f)

Joint work with Cyril Cohen, LMCS 2012.

⇒ A (constructive) theory of semi-algebraic sets and functions.Boris Djalal, Proceedings of CPP 2018.

Assia Mahboubi – Verified Computer-Aided Mathematics 47

Page 89: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Formal abstracts?

Uniform p-adic cell decomposition and local zeta functions,Johan Pas, J. reine angew. Math. 399 (1989), 137—172.

Define:

qelim : formula -> formula

Kq_free : formula -> B

Such that:

forall f : formula, Kq_free (qelim f)

Now:

forall f : formula,

exists out : {finset N}, forall (p : N), prime p -> p /∈ out ->

forall (e : seq Qp), Qp, holds Qp e f <-> holds Qp e (qelim f).

Assia Mahboubi – Verified Computer-Aided Mathematics 48

Page 90: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Formal abstracts?

Uniform p-adic cell decomposition and local zeta functions,Johan Pas, J. reine angew. Math. 399 (1989), 137—172.

Define:

qelim : formula -> formula

Kq_free : formula -> B

Such that:

forall f : formula, Kq_free (qelim f)

Now:

forall f : formula,

exists out : {finset N}, forall (p : N), prime p -> p /∈ out ->

forall (e : seq Qp), Qp, holds Qp e f <-> holds Qp e (qelim f).

Assia Mahboubi – Verified Computer-Aided Mathematics 48

Page 91: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Formal abstracts?

Uniform p-adic cell decomposition and local zeta functions,Johan Pas, J. reine angew. Math. 399 (1989), 137—172.

Define:

qelim : formula -> formula

Kq_free : formula -> B

Such that:

forall f : formula, Kq_free (qelim f)

Now:

forall f : formula,

exists out : {finset N}, forall (p : N), prime p -> p /∈ out ->

forall (e : seq Qp), Qp, holds Qp e f <-> holds Qp e (qelim f).

Assia Mahboubi – Verified Computer-Aided Mathematics 48

Page 92: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Formal abstracts?

Uniform p-adic cell decomposition and local zeta functions,Johan Pas, J. reine angew. Math. 399 (1989), 137—172.

Define:

qelim : formula -> formula

Kq_free : formula -> B

Such that:

forall f : formula, Kq_free (qelim f)

Now:

forall f : formula,

exists out : {finset N}, forall (p : N), prime p -> p /∈ out ->

forall (e : seq Qp), Qp, holds Qp e f <-> holds Qp e (qelim f).

Assia Mahboubi – Verified Computer-Aided Mathematics 48

Page 93: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Formal abstracts?

Uniform p-adic cell decomposition and local zeta functions,Johan Pas, J. reine angew. Math. 399 (1989), 137—172.

Define:

qelim : formula -> formula

Kq_free : formula -> B

Such that:

forall f : formula, Kq_free (qelim f)

Now:

forall f : formula,

exists out : {finset N}, forall (p : N), prime p -> p /∈ out ->

forall (e : seq Qp), Qp, holds Qp e f <-> holds Qp e (qelim f).

Assia Mahboubi – Verified Computer-Aided Mathematics 48

Page 94: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Conclusion

Happy Birthday Tom

Assia Mahboubi – Verified Computer-Aided Mathematics 49

Page 95: Veri edComputer-Aided Mathematics[Introduction to Interval Analysis - R. E. Moore, R. B. Kearfott, M. J. Cloud] [Validated Numerics: A Short Introduction to Rigorous Computations -

Conclusion

Happy Birthday Tom

Assia Mahboubi – Verified Computer-Aided Mathematics 49