SC Chapter 8 Quantifier-free Non-linear Real Arithmetic

Questions about the lecture 'Satisfiability Checking' of the RWTH Aachen Chapter 8 Quantifier-free Non-linear Real Arithmetic

Questions about the lecture 'Satisfiability Checking' of the RWTH Aachen Chapter 8 Quantifier-free Non-linear Real Arithmetic


R. B.
This flashcard set covers advanced topics in quantifier-free non-linear real arithmetic at the university level. It delves into concepts like polynomial constraints, propagation, and the Cauchy bound, focusing on the characteristics and properties of polynomials, roots, and regions. The set also explores methods like the Sturm sequence, virtual substitution, and the Newton method, which are crucial for solving and understanding non-linear arithmetic problems. This flashcard set is particularly beneficial for university students and researchers in computer science who need to grasp the intricacies of real arithmetic and its applications in algorithmic problem-solving.
Cartes-fiches
63
Utilisateurs
1
Langue
Anglais
Catégorie
Informatique
Niveau
Université
Créé / Mis à jour
18.02.2017 / 28.02.2017

Cartes-fiches

What are the components of the syntax of nonlinear real arithmetics? [3]

1. Polynomials p 2. constraints c and 3. formulas phi

What are the members of a polynomials p of nonlinear real arithmetics? [5]

const, variable x, p+p, p–p, p*p // * makes the difference to linear

What are the members of a constraint c of nonlinear real arithmetics? [2]

p < 0, p = 0

What are the members of a formula phi of nonlinear real arithmetics? [3]

c, \neg phi, phi \land phi

What is a characteristic of interval constraint propagation (ICP)? [3]

1. Incomplete, might return “unknown”

2. Combine with complete procedure implemented in the backend

 

3. Cheap reduction of search space method in case of “unknown”

What is the definition of an interval I? [5]

1. Left and right including or excluding (with inf) bounds

2. Simple case with [l,u], [l,inf) and (-inf,u] only

3. Bounded if both bounds are real-valued

 

4. Diameter is width of interval

5. n-dimensional box with [l1,u1] x … x [ln,un]

 

What are the rules for interval arithmetic? [7]

1. Constants and variables are interval-valued

 

2. A+B = [la+lb, ua+ub]

3. A-B = [la-ub, ua-lb]

4. A*B = [min_prod, max_prod]

5. A*A = [min_prod, max_prod] cut [0,inf)

6. A/B = A*1/B // 0 \notin B

7. A/B = (-inf,1/min] cup [1/max,inf) // 0 \in B

What is the definition of a contraction candidate (CC)?

Pair of constraint and variable

What are the characteristics of a contraction operator? [3]

1. Not dropping solutions

2. Does current box has no solution? // Usually not guaranteed

3. Contract as long as gain is large enough

What are possible contraction operators? [2]

1. Interval constraint propagation (ICP)

2. Interval Newton method (INM)

What is the goal of the interval constraint propagation algorithm?

Goal is to obtain an empty box (UNSAT) or to reach a diameter

What is the solution of the interval constraint propagation algorithm?

Solution is a non-empty box forwarded to a completing method

What is the step of the interval constraint propagation algorithm?

Choose CC and perform a contraction

 

What are the characteristics of the weight heuristic for the next contraction in the interval constraint propagation algorithm? [5]

1. rk+1ij = gainrel = 1 – Dnew/Dold

2. Each contraction has weight wkij \in [0,1]

3. Update weight with wk+1ij = wkij + a*(rk+1ij – wkij)

4. Choose CC with highest weight over threshold eps

 

5. If no CC is over threshold eps split manually

What is another heuristic beside the weight heuristic for the next contraction in the interval constraint propagation algorithm?

Traverse history tree pre-order

What are the characteristics of the history tree for the interval constraint propagation algorithm? [4]

1. Search box

2. Applied contractions

3. Splitting dimensions

 

4. Collect infeasible subsets // Also possible

What are possible improvements of the interval constraint propagation algorithm? [5]

1. More sophisticated contraction (INM)

2. Linear solver for linear constraints

3. Box validation against linear feasible regions // Separate inside from partly covered case

4. SAT-solver for box-choice

5. Splitting heuristics

How complex is satisfiability checking for QFNRA?

Best known method for QFNRA formulas have exponential complexity

What holds for the diameter D of an interval? [4]

1. D_i = u-l

2. inf if i unbounded

3. 0 if i apoint interval

4. negative if i empty

How does the preprocessing for ICP contraction works?

Linearization, separate linear and nonlinear constrains with new variables

How to gain new bounds for ICP contraction? [3]

1. Update via cut with old bound

2. If < or <= take lold and unew

3. If > or >= take lnew and uold

What are the two steps of the ICP contraction? [2]

1. Insert current bounds for all except one variable in one constraint

2. Choose box if split of intervals occurred

How does the preprocessing for INM contraction works? [2]

1. Transform e1 ~ e2 to e1-e2 ~ 0

2. Transform inequation p ~ 0 to h-p = 0

How to gain new bounds for INM contraction? [3]

1. Update via cut with old bound

2. If < or <= take lold and min of uold or 0

3. If > or >= take max of lold or 0 and uold

What are the two steps of the INM contraction? [3]

1. Set starting interval A and sample point s

2. Compute A’ with Newton method

3. Get new bound

What is the original idea of the Newton method? [3]

1. Make initial guess so for univariate polynomial f(x)

2. si+1 = si – f(si)/f’(si)

3. Usually converges quadratically to r with f(r) = 0

What is the original idea of the Taylor’s theorem? [2]

1. For all s,x\in A exists d\in A

2. With f(x) = f(s) + (x-s)*f’(d)

What is the original idea of the componentwise multivariate Newton?

si+1 = Ncmp(s,f(x),xj) = si – f(si)/[df/dxj](si)

What is the adapted idea of the Newton method?

s – f(s)/f’(A) with s\in A

What is the adapted idea of the componentwise multivariate Newton? [6]

1. A = A1 x … x An with mid(Aj) middle point of interval

2. si+1 = Ncmp(A,f(x),xj) = mid(Aj)- f(A1,…,An)/[df/dxj](A1,…,An)

 

3. If f(x)=0 then x is in Ncmp

4. If cut is empty no x yields to f(x) = 0

 

5. No diverging behavior as the original

6. Drop boxes if they contract to empty

What kind of procedure is virtual substitution?

Existential quantifier elimination procedure

What is the condition of virtual substitution?

For every constraint of formula p(x)~0 degree of x in p(x) <= 2

What are the steps of virtual substitution? [2]

1. \exists is replaced by t \in test candidates T [t//var]

2. Side conditions are added

What is the goal of virtual substitution?

Goal is to find and return satisfying solution or explain unsatisfiability

What are the finite endpoints cases for virtual substitution for p=ax²+bx+c? [4]

1. If p is constant (for a=b=0) or no zeros then sign-invariant

2. If a=0 and b!=0 then x0 = -c/b

3. If a!=0 and b²-4ac>=0 then x1 = -b+sqrt(b²-4ac)/2a

4. If a!=0 and b²-4ac>0 then x2 = -b-sqrt(b²-4ac)/2a

What are the solution intervals cases for virtual substitution? [4]

1. If p=0 then [xi,xi]

2. If p < > != 0 then (-inf,xi), (xi,xj) and (xi,inf)

3. If p <= >= 0 then (-inf,xi), [xi,xi], [xi,xj] and (xi,inf)

4. Always (-inf,inf)

What are the test candidates T cases for virtual substitution? [2]

1. If p <= >= = 0 then zeros of p and -inf (A sufficiently small value)

2. If p < > != 0 then zeros of p+eps and -inf

What are the types of test candidates for virtual substitution? [3]

1. -inf

2. +eps

3. Contains sqrt

What are the forms of polynomials?

[definition.CAD.QFNRA, 2]

1. Normal and 2. univariate

What is the definition?

[normal form.definition.CAD.QFNRA]

p = a1x1e11 … xnen1 + … + akx1e1k … xnenk

Étudier