Direct Methods for Sparse Linear Systems Lecture 4 Alessandra Nardi Thanks to Prof. Jacob White, Suvranu De, Deepak Ramaswamy, Michal Rewienski, and Karen

Direct Methods for Sparse Linear Systems

Lecture 4

Alessandra Nardi

Thanks to Prof. Jacob White, Suvranu De, Deepak Ramaswamy, Michal Rewienski, and Karen Veroy

Last lecture review

• Solution of system of linear equations

• Existence and uniqueness review

• Gaussian elimination basics– GE basics– LU factorization– Pivoting

Outline

• Error Mechanisms

• Sparse Matrices – Why are they nice– How do we store them– How can we exploit and preserve sparsity

Error Mechanisms

• Round-off error– Pivoting helps

• Ill conditioning (almost singular)– Bad luck: property of the matrix– Pivoting does not help

• Numerical Stability of Method

Ill-Conditioning : Norms

• Norms useful to discuss error in numerical problems

• Norm V:

Vyxyxyx

Vxxx

Vxxx

, if )3(

, if )2(

,0 if 0 )1(

L2 (Euclidean) norm :

n

iixx

1

2

2

L1 norm :

L norm :

n

iixx

11

ii

xx max

12x

Unit circle

Unit square

1

1 11x

1

x

Ill-Conditioning : Vector Norms

Vector induced norm : Axx

AxA

xx 1maxmax

n

iij

jAA

11

max

Induced norm of A is the maximum “magnification” of by

= max abs column sum

n

jij

iAA

1

max = max abs row sum

2A = (largest eigenvalue of ATA)1/2

x A

Ill-Conditioning : Matrix Norms

Ill-Conditioning : Matrix Norms

• More properties on the matrix norm:

• Condition Number:

-It can be shown that:-Large (A) means matrix is almost singular (ill-conditioned)

BAAB

I

1

)( 1 AAA

1)( A

What happens if we perturb b?

Ill-Conditioning: Perturbation Analysis

xxxbbb

b

bM

x

x

b

bMM

x

x

xMb

bM

b

bbMx

bMx

bMx

bxM

bbxMMx

bbxxM

)(

)(

1

11

1

1

(M) large is bad

What happens if we perturb M?


xxxMMM

M

MM

xx

x

M

MMM

xx

x

bxxMM

)(

))((

1

(M) large is bad

Bottom line:If matrix is ill-conditioned, round-off puts you in troubles

Geometric Approach is more intuitive

1 2 1 1 2 2[ ], Solving is finding M M M M x b x M x M b

2

2|| ||

M

M

1

1|| ||

M

M

2

2|| ||

M

M

1

1|| ||

M

M

When vectors are nearly aligned,Hard to decide how much of versus how much of 1M

2M

Vectors are orthogonal

1x

2x

1x

2x


Vectors are nearly aligned

Numerical Stability

• Rounding errors may accumulate and propagate unstably in a bad algorithm.

• Can be proven that for Gaussian elimination the accumulated error is bounded

Summary on Error Mechanisms for GE

• Rounding: due to machine finite precision we have an error in the solution even if the algorithm is perfect– Pivoting helps to reduce it

• Matrix conditioning– If matrix is “good”, then complete pivoting solves any round-

off problem– If matrix is “bad” (almost singular), then there is nothing to

do

• Numerical stability– How rounding errors accumulate– GE is stable

LU – Computational Complexity

• Computational Complexity– O(n3), where M: n x n

• We cannot afford this complexity

• Exploit natural sparsity that occurs in circuits equations– Sparsity: many zero elements– Matrix is sparse when it is advantageous to exploit

its sparsity

• Exploiting sparsity: O(n1.1) to O(n1.5)

LU – Goals of exploiting sparsity

(1) Avoid storing zero entries– Memory usage reduction– Decomposition is faster since you do need to access

them (but more complicated data structure)

(2) Avoid trivial operations– Multiplication by zero– Addition with zero

(3) Avoid losing sparsity

1 2 3 4 1m m

X X

X X X

X X X

X X X

X X X

X X X

X X X

X X

m

Sparse Matrices – Resistor Line

Tridiagonal Case

For i = 1 to n-1 { “For each Row”

For j = i+1 to n { “For each target Row below the source”

For k = i+1 to n { “For each Row element beyond Pivot”

} }}

Pivotjiji

ii

MM

M

Multiplier

jk jk ji ikM M M M

Order N Operations!

GE Algorithm – Tridiagonal Example

1R

5R

3R

4R

2R1V 2V

3V

1Si

1

1

R 2

1

R

2

1

R

2

1

R

2

1

R 3

1

R

4

1

R

4

1

R

4

1

R

4

1

R 5

1

R

SymmetricDiagonally Dominant

Nodal Matrix0

Sparse Matrices – Fill-in – Example 1

13

2

1

0

0

sie

e

e

X X X

X X 0

X 0 X

X X X

X X 0

X 0 X

X

X

X X

X= Non zero

Matrix Non zero structure Matrix after one LU step

X X


X X X X

X X 0 0

0 X X 0

X 0 00

Fill-ins Propagate

XX

X

X

X

X X

X

X X

Fill-ins from Step 1 result in Fill-ins in step 2


3V

Node Reordering Can Reduce Fill-in - Preserves Properties (Symmetry, Diagonal Dominance) - Equivalent to swapping rows and columns

1V 2V

0

x x x

x x x

x x x

Fill-ins

2V 1V

3V

0

x x 0

x x x

0 x x

No Fill-ins

Sparse Matrices – Fill-in & Reordering

Exploiting and maintaining sparsity• Criteria for exploiting sparsity:

– Minimum number of ops– Minimum number of fill-ins

• Pivoting to maintain sparsity: NP-complete problem heuristics are used– Markowitz, Berry, Hsieh and Ghausi, Nakhla and Singhal and

Vlach– Choice: Markowitz

• 5% more fill-ins but• Faster

• Pivoting for accuracy may conflict with pivoting for sparsity

Where can fill-in occur ?

x x x

x x x

x x x

Multipliers

Already Factored

Possible Fill-inLocations

Fill-in Estimate = (Non zeros in unfactored part of Row -1) (Non zeros in unfactored part of Col -1) Markowitz product


Markowitz Reordering (Diagonal Pivoting)

tor oF 1i n diagonal with min Markowitz PFind roductj i

rows and columns Swa p ij j i

the new row and determine fiFactor ll-insi

End

Greedy Algorithm (but close to optimal) !


Why only try diagonals ?

• Corresponds to node reordering in Nodal formulation

03

1 2

02

3 1

• Reduces search cost • Preserves Matrix Properties - Diagonal Dominance - Symmetry


Pattern of a Filled-in Matrix

Very Sparse

Very Sparse

Dense



Unfactored random Matrix


Factored random Matrix

Sparse Matrices – Data Structure

• Several ways of storing a sparse matrix in a compact form

• Trade-off– Storage amount– Cost of data accessing and update procedures

• Efficient data structure: linked list

Sparse Matrices – Data Structure 1

Orthogonal linked list

Val 11

Col 11

Val 12

Col 12

Val 1K

Col 1K

Val 21

Col 21

Val 22

Col 22

Val 2L

Col 2L

Val N1

Col N1

Val N2

Col N2

Val Nj

Col Nj

1

N

Arrays of Data in a RowVector of row pointers

Matrix entriesColumn index

Sparse Matrices – Data Structure 2

, 1 , 4 , 5 , 7 , 9 , 12 , 15

1 4 5 7 9 12 15i i i i i i i i i i i i i iM M M M M M M

i i i i i i i

, 1 , 7 , 15

1 7 15i i i i i iM M M

i i i

Eliminating Source Row i from Target row j

Row i

Row j

Must read all the row j entries to find the 3 that match row iEvery Miss is an unneeded memory reference (expensive!!!)

Could have more misses than ops!

Sparse Matrices – Data StructureProblem of Misses

, 1 , 4 , 5 , 7 , 9 , 12 , 15

1 4 5 7 9 12 15i i i i i i i i i i i i i iM M M M M M M

i i i i i i i

Row j

, 1i iM , 4i iM , 5i iM , 7i iM , 9i iM , 12i iM , 15i iM

1) Read all the elements in Row j, and scatter them in an n-length vector2) Access only the needed elements using array indexing!

Sparse Matrices – Data Structure

Scattering for Miss Avoidance

X

X

X

X

X

X

X X

X

XX

X

X

X X 1

2

4

3

5

• One Node Per Matrix Row

• One Edge Per Off-diagonal Pair

X

X

Structurally Symmetric Matrices and Graphs

Sparse Matrices – Graph Approach

X

X

X

X

X

X

X X

X

XX

X

X

X X

X

X 1

2

4

3

5

Markowitz Products = (Node Degree)2

2(degree 1) 92(deg ree 2) 42(deg ree 3) 9

2(degree 4) 42(degree 5) 4

Sparse Matrices – Graph ApproachMarkowitz Products

93311 M42222 M93333 M42244 M42255 M

X

X

X

X

X

X

X X

X

XX

X

X

X X 1

2

4

3

5

• Delete the node associated with pivot row

• “Tie together” the graph edges

X

X

X

X

X

One Step of LU Factorization

Sparse Matrices – Graph ApproachFactorization

x x x x

x x x

x x x x

x x x x

x x x x

1 2 3 4 5

Graph

Markowitz products( = Node degree)

Col Row

3 3 9

2 2 4

3 3 9

3 3 9

3 3 9

1

2

3

4

5

Sparse Matrices – Graph ApproachExample

x x x

x x x x x

x x x x x

x x x x

x x x x

1 3 4 5

Graph

Swap 2 with 1

Sparse Matrices – Graph ApproachExample

• Gaussian Elimination Error Mechanisms– Ill-conditioning

– Numerical Stability

• Gaussian Elimination for Sparse Matrices – Improved computational cost: factor in O(N1.5)

operations (dense is O(N3) )

– Example: Tridiagonal Matrix Factorization O(N)

– Data structure

– Markowitz Reordering to minimize fill-ins

– Graph Based Approach

Summary

Documents

Direct Methods for Sparse Linear Systems Lecture 4 Alessandra Nardi Thanks to Prof. Jacob White, Suvranu De, Deepak Ramaswamy, Michal Rewienski, and Karen