an exact toric resultant-based rur approach for solving polynomial systems

Post on 19-Jan-2016

29 Views

Category:

Documents

0 Downloads

Preview:

Click to see full reader

DESCRIPTION

An Exact Toric Resultant-Based RUR Approach for Solving Polynomial Systems. Koji Ouchi, John Keyser, J. Maurice Rojas Department of Computer Science, Mathematics Texas A&M University AMS Meeting 2004. Outline. Rational Univariate Reduction (RUR) Complexity Analysis Exact RUR - PowerPoint PPT Presentation

TRANSCRIPT

An Exact Toric Resultant-BasedRUR Approach

for Solving Polynomial Systems

Koji Ouchi, John Keyser, J. Maurice Rojas

Department of Computer Science, Mathematics

Texas A&M University

AMS Meeting 2004

Texas A&M University AMS2004 2RUR

Outline

Rational Univariate Reduction (RUR) Complexity Analysis Exact RUR Comparison with Other Work Conclusion / Future Work

Texas A&M University AMS2004 3RUR

Outline

Rational Univariate Reduction (RUR) Complexity Analysis Exact RUR Comparison with Other Work Conclusion / Future Work

Texas A&M University AMS2004 4RUR

Rational Univariate Reduction

Problem: Solve a system of n polynomials f1, …, fn

in n variables X1, …, Xn

with coefficients in ℚℚ

Reduce the system to

n + 1 univariate polynomials h, h1, …, hn

with coefficients in ℚℚ s.t.

if is a root of h then

(h1(), …, hn()) is a solution to the system

Texas A&M University AMS2004 5RUR

Notation u = (u0, u1,…, un) indeterminates f0 = u0 + u1 X1 + … + un Xn

Ai = Supp(fi), i = 0, 1,…, n ∴ A0 = {o, e1, …, en}

ei the i-th standard basis vector

RUR via Toric Resultant

Texas A&M University AMS2004 6RUR

Toric Perturbation

Toric Generalized Characteristic PolynomialLet f1

*, …, fn* be n polynomials

in n variables X1, …, Xn

with coefficients in ℚ ℚ andSupp(fi

*) ⊆ Ai = Supp(fi ), i = 0, 1,…, nthat have only finitely many solutions in (ℂℂ \ {0})n

DefineTGCP(u, Y ) =

Res (A0, A1, …, An) (f0, f1 - Y f1*, …, fn - Y fn

*)

Texas A&M University AMS2004 7RUR

Toric Perturbation

Toric Perturbation [Rojas 99]Define Pert(u) to be

the non-zero coefficient of the lowest degree term(in Y ) of TGCP(u, Y )

Pert(u) is well-defined A version of “perturbations” [D’Andrea and Emiris

01, 03]

Texas A&M University AMS2004 8RUR

Toric Perturbation

Toric Perturbation If (1, …, n) (ℂℂ \ {0})n is an isolated root of

the input system f1, …, fn then u0 + u1 1 + … + un n Pert(u)

Pert(u) completely splits into linear factorsover ℂℂ

For every irreducible component of the zero setof the input system, there is at least one factor ofPert(u)

Texas A&M University AMS2004 9RUR

Computing RUR Step1: Compute Mixed Volumes Step2: Construct a Resultant Matrix Step3: Compute h Step4: Compute h1, …, hn

Texas A&M University AMS2004 10RUR

Computing RUR Step 1: Compute Mixed Volumes

Use Emiris’s algorithm [Emiris and Canny 95, 01] to compute

MV–i = MV(A0, A1, …, Ai-1, Ai+1, …, An), i = 0, 1, …, n

Use Linear Programming #P on Turing machine

1. Mixed Volumes2. Resultant Matrix3. h4. h1, …, hn

Texas A&M University AMS2004 11RUR

Computing RUR Step 2: Construct a Resultant Matrix

Use Emiris’ algorithm [Emiris and Canny 95]to construct a matrixwhose maximal minor is some multiple ofthe toric resultant

Rows and columns are labeledby the exponents in A0, A1, …, An

Increment rows and columns until non-vanishing maximal minor is found

1. Mixed Volumes2. Resultant Matrix3. h4. h1, …, hn

Texas A&M University AMS2004 12RUR

Computing RUR Step 2: Construct a Resultant Matrix (Cont.)

[Pederson and Sturmfels 93]deg fi Res (A0, A1, …, An) (f0, f1, …, fn) = MV-i , i = 0, 1,…, n

1. Mixed Volumes2. Resultant Matrix3. h4. h1, …, hn

Texas A&M University AMS2004 13RUR

Computing RUR

Step 2: Construct a Resultant Matrix (Cont.) Degeneracies have been removed by perturbation

The size of matrices must be at least Σ i = 0, 1,…, n MV-i

# of rows labeled by the exponents in Ai ≧ MV-i , i = 0, 1, …, n

# of rows labeled by the exponents in A0 = MV-0

∴ deg f0 D = MV-0

where D is the maximal minor

1. Mixed Volumes2. Resultant Matrix3. h4. h1, …, hn

Texas A&M University AMS2004 14RUR

Computing RUR

Step 3: Compute h (T)

h(T) = Pert(T, u1, …, un) for some values of u1, …, un

Assign values to u1, …, un Evaluate Pert(u0, u1, …, un)

at deg h(T) = MV-0 distinct values of u0 andinterpolate them

1. Mixed Volumes2. Resultant Matrix3. h4. h1, …, hn

Texas A&M University AMS2004 15RUR

Computing RUR

Step 4: Compute h1 (T), …, hn (T)

Computation of every hi involves Evaluating Pert(u) and interpolate them Univariate polynomial operations

Euclidean algorithm for GCD First subresultant [Gonzalez-Vega 91]

1. Mixed Volumes2. Resultant Matrix3. h4. h1, …, hn

Texas A&M University AMS2004 16RUR

Computing RUR All the steps can be implemented exactly

The coefficients of h, h1, …, hn can be computedin full digits

Texas A&M University AMS2004 17RUR

Outline

Rational Univariate Reduction (RUR) Complexity Analysis Exact RUR Comparison with Other Work Conclusion / Future Work

Texas A&M University AMS2004 18RUR

Complexity Analysis

Notation O˜( ) the polylog factor is ignored

Gaussian elimination ofm dimensional matrix requiresO(m) operations

Texas A&M University AMS2004 19RUR

Complexity Analysis Quantities

MA MV-0 = deg h(T)

RA i = 0, 1,…, n MV-i

The size of the optimal resultant matrix

SA The size of maximal minor SA = (n1/2 en RA)

Texas A&M University AMS2004 20RUR

Complexity Analysis

# of Arithmetic Operations Evaluate Res (A0, A1, …, An) O˜(SA

1+) Evaluate Pert (u) O˜(SA

1+) Compute h O˜(MA SA

1+) Compute every hi O˜(MA SA

1+) Compute RUR

for fixed u1, …, un O˜(n MA SA1+)

Compute RUR O˜(n3 MA3 SA

1+)

Texas A&M University AMS2004 21RUR

Complexity Analysis

Bit Complexity The logarithmic height of h, h1, …, hn is

some polynomial in SA [Rojas 00] RA [Sombra]

The bit complexity is single exponential in n

Texas A&M University AMS2004 22RUR

Complexity Analysis A great speed up is achieved

if we could compute “small” matrixwhose determinant is the resultant No such method is known

Resultant matrices Sylvester-Dixon [Chtcherba and Kapur] Corner-cutting [Goldman and Zhang 00] Tate resolution [Khetan 03, 04]

Texas A&M University AMS2004 23RUR

Khetan’s Method Khetan’s method gives a matrix

whose determinant is the resultantof unmixed systems when n = 2 or 3 (or bigger?)[Khetan 03, 04]

Let B = A0 A1 An

Then, computing RUR requires

n3 MA3 RB

1+

arithmetic operations

Texas A&M University AMS2004 24RUR

Outline

Rational Univariate Reduction (RUR) Complexity Analysis Exact RUR Comparison with Other Work Conclusion / Future Work

Texas A&M University AMS2004 25RUR

ERUR Non square system is converted to

some square system

Solutions in ℂℂn are computedby adding the origin o to supports

In both cases,post processing requires exact computationover the points in RUR

Texas A&M University AMS2004 26RUR

ERUR

Exact Sign Given an expression e, tell whether or not

e(h1(), …, hn()) = 0

Use (extended) root bound approach Use Aberth’s method [Aberth 73] to

numerically compute an approximation fora root of hto any precision

Im1Re

Texas A&M University AMS2004 27RUR

Applications by ERUR

Real Root Given a system of polynomial equations,

list all the real roots of the system

Positive Dimensional Component Given a system of polynomial equations,

tell whether or not the zero set of the systemhas a positive dimensional component

Texas A&M University AMS2004 28RUR

Outline

Rational Univariate Reduction (RUR) Complexity Analysis Exact RUR Comparison with Other Work Conclusion / Future Work

Texas A&M University AMS2004 29RUR

The Other RUR

GB+RS [Rouillier 99, 04]

Kronecker / Newton [Giusti, Lecerf and Salvy 01] [Jeronimo, Krick, Sabia and Sombra 04]

Texas A&M University AMS2004 30RUR

The Other RUR

GB+RS [Rouillier 99, 04] Compute the exact RUR for real solutions

of a 0-dimensional system

GB computes the Gröbner basis The Gröbner basis computation is

EXPSPACE-complete (double exponential in n)on Turing machine [Mayr and Meyer 98]

Texas A&M University AMS2004 31RUR

The Other RUR

Kronecker / Newton [Giusti, Lecerf and Salvy 01]

Kronecker in Magma

[Jeronimo, Krick, Sabia and Sombra 04] BPP on BSS machine over ℚℚ

Texas A&M University AMS2004 32RUR

Outline

Rational Univariate Reduction (RUR) Complexity Analysis Exact RUR Comparison with Other Work Conclusion / Future Work

Texas A&M University AMS2004 33RUR

Implementation

ERUR Algorithms adapt to exact implementation naturally

Strong for handling degeneracies

Need more optimizations and faster algorithms

Texas A&M University AMS2004 34RUR

Conclusion

Deterministic algorithm

Handle degeneracies by perturbation The total degree of Pert(u) is RA

Use the incremental matrix construction algorithm Currently, the most efficient Starting at a matrix of size RA

Exponential factor appearing in the complexitycomes from the size of the resultant matrix

Texas A&M University AMS2004 35RUR

Future Work Faster toric resultant algorithms

Smaller resultant matrices

Take advantages of sparseness of matrices[Emiris and Pan 97]

Faster univariate polynomial operations Use rational functions for h1,…, hn

Texas A&M University AMS2004 36RUR

Thank you for listening!

Contact Koji Ouchi, kouchi@cs.tamu.edu John Keyser, keyser@cs.tamu.edu Maurice Rojas, rojas@math.tamu.edu

Visit Our Web http://research.cs.tamu.edu/keyser/geom/ERUR/

Thank you

top related