| 1 |
<?php |
| 2 |
|
| 3 |
namespace Matrix\Decomposition; |
| 4 |
|
| 5 |
use Matrix\Exception; |
| 6 |
use Matrix\Matrix; |
| 7 |
|
| 8 |
class LU |
| 9 |
{ |
| 10 |
private $luMatrix; |
| 11 |
private $rows; |
| 12 |
private $columns; |
| 13 |
|
| 14 |
private $pivot = []; |
| 15 |
|
| 16 |
public function __construct(Matrix $matrix) |
| 17 |
{ |
| 18 |
$this->luMatrix = $matrix->toArray(); |
| 19 |
$this->rows = $matrix->rows; |
| 20 |
$this->columns = $matrix->columns; |
| 21 |
|
| 22 |
$this->buildPivot(); |
| 23 |
} |
| 24 |
|
| 25 |
/** |
| 26 |
* Get lower triangular factor. |
| 27 |
* |
| 28 |
* @return Matrix Lower triangular factor |
| 29 |
*/ |
| 30 |
public function getL() |
| 31 |
{ |
| 32 |
$lower = []; |
| 33 |
|
| 34 |
$columns = min($this->rows, $this->columns); |
| 35 |
for ($row = 0; $row < $this->rows; ++$row) { |
| 36 |
for ($column = 0; $column < $columns; ++$column) { |
| 37 |
if ($row > $column) { |
| 38 |
$lower[$row][$column] = $this->luMatrix[$row][$column]; |
| 39 |
} elseif ($row === $column) { |
| 40 |
$lower[$row][$column] = 1.0; |
| 41 |
} else { |
| 42 |
$lower[$row][$column] = 0.0; |
| 43 |
} |
| 44 |
} |
| 45 |
} |
| 46 |
|
| 47 |
return new Matrix($lower); |
| 48 |
} |
| 49 |
|
| 50 |
/** |
| 51 |
* Get upper triangular factor. |
| 52 |
* |
| 53 |
* @return Matrix Upper triangular factor |
| 54 |
*/ |
| 55 |
public function getU() |
| 56 |
{ |
| 57 |
$upper = []; |
| 58 |
|
| 59 |
$rows = min($this->rows, $this->columns); |
| 60 |
for ($row = 0; $row < $rows; ++$row) { |
| 61 |
for ($column = 0; $column < $this->columns; ++$column) { |
| 62 |
if ($row <= $column) { |
| 63 |
$upper[$row][$column] = $this->luMatrix[$row][$column]; |
| 64 |
} else { |
| 65 |
$upper[$row][$column] = 0.0; |
| 66 |
} |
| 67 |
} |
| 68 |
} |
| 69 |
|
| 70 |
return new Matrix($upper); |
| 71 |
} |
| 72 |
|
| 73 |
/** |
| 74 |
* Return pivot permutation vector. |
| 75 |
* |
| 76 |
* @return Matrix Pivot matrix |
| 77 |
*/ |
| 78 |
public function getP() |
| 79 |
{ |
| 80 |
$pMatrix = []; |
| 81 |
|
| 82 |
$pivots = $this->pivot; |
| 83 |
$pivotCount = count($pivots); |
| 84 |
foreach ($pivots as $row => $pivot) { |
| 85 |
$pMatrix[$row] = array_fill(0, $pivotCount, 0); |
| 86 |
$pMatrix[$row][$pivot] = 1; |
| 87 |
} |
| 88 |
|
| 89 |
return new Matrix($pMatrix); |
| 90 |
} |
| 91 |
|
| 92 |
/** |
| 93 |
* Return pivot permutation vector. |
| 94 |
* |
| 95 |
* @return array Pivot vector |
| 96 |
*/ |
| 97 |
public function getPivot() |
| 98 |
{ |
| 99 |
return $this->pivot; |
| 100 |
} |
| 101 |
|
| 102 |
/** |
| 103 |
* Is the matrix nonsingular? |
| 104 |
* |
| 105 |
* @return bool true if U, and hence A, is nonsingular |
| 106 |
*/ |
| 107 |
public function isNonsingular() |
| 108 |
{ |
| 109 |
for ($diagonal = 0; $diagonal < $this->columns; ++$diagonal) { |
| 110 |
if ($this->luMatrix[$diagonal][$diagonal] === 0.0) { |
| 111 |
return false; |
| 112 |
} |
| 113 |
} |
| 114 |
|
| 115 |
return true; |
| 116 |
} |
| 117 |
|
| 118 |
private function buildPivot() |
| 119 |
{ |
| 120 |
for ($row = 0; $row < $this->rows; ++$row) { |
| 121 |
$this->pivot[$row] = $row; |
| 122 |
} |
| 123 |
|
| 124 |
for ($column = 0; $column < $this->columns; ++$column) { |
| 125 |
$luColumn = $this->localisedReferenceColumn($column); |
| 126 |
|
| 127 |
$this->applyTransformations($column, $luColumn); |
| 128 |
|
| 129 |
$pivot = $this->findPivot($column, $luColumn); |
| 130 |
if ($pivot !== $column) { |
| 131 |
$this->pivotExchange($pivot, $column); |
| 132 |
} |
| 133 |
|
| 134 |
$this->computeMultipliers($column); |
| 135 |
|
| 136 |
unset($luColumn); |
| 137 |
} |
| 138 |
} |
| 139 |
|
| 140 |
private function localisedReferenceColumn($column) |
| 141 |
{ |
| 142 |
$luColumn = []; |
| 143 |
|
| 144 |
for ($row = 0; $row < $this->rows; ++$row) { |
| 145 |
$luColumn[$row] = &$this->luMatrix[$row][$column]; |
| 146 |
} |
| 147 |
|
| 148 |
return $luColumn; |
| 149 |
} |
| 150 |
|
| 151 |
private function applyTransformations($column, array $luColumn) |
| 152 |
{ |
| 153 |
for ($row = 0; $row < $this->rows; ++$row) { |
| 154 |
$luRow = $this->luMatrix[$row]; |
| 155 |
// Most of the time is spent in the following dot product. |
| 156 |
$kmax = min($row, $column); |
| 157 |
$sValue = 0.0; |
| 158 |
for ($kValue = 0; $kValue < $kmax; ++$kValue) { |
| 159 |
$sValue += $luRow[$kValue] * $luColumn[$kValue]; |
| 160 |
} |
| 161 |
$luRow[$column] = $luColumn[$row] -= $sValue; |
| 162 |
} |
| 163 |
} |
| 164 |
|
| 165 |
private function findPivot($column, array $luColumn) |
| 166 |
{ |
| 167 |
$pivot = $column; |
| 168 |
for ($row = $column + 1; $row < $this->rows; ++$row) { |
| 169 |
if (abs($luColumn[$row]) > abs($luColumn[$pivot])) { |
| 170 |
$pivot = $row; |
| 171 |
} |
| 172 |
} |
| 173 |
|
| 174 |
return $pivot; |
| 175 |
} |
| 176 |
|
| 177 |
private function pivotExchange($pivot, $column) |
| 178 |
{ |
| 179 |
for ($kValue = 0; $kValue < $this->columns; ++$kValue) { |
| 180 |
$tValue = $this->luMatrix[$pivot][$kValue]; |
| 181 |
$this->luMatrix[$pivot][$kValue] = $this->luMatrix[$column][$kValue]; |
| 182 |
$this->luMatrix[$column][$kValue] = $tValue; |
| 183 |
} |
| 184 |
|
| 185 |
$lValue = $this->pivot[$pivot]; |
| 186 |
$this->pivot[$pivot] = $this->pivot[$column]; |
| 187 |
$this->pivot[$column] = $lValue; |
| 188 |
} |
| 189 |
|
| 190 |
private function computeMultipliers($diagonal) |
| 191 |
{ |
| 192 |
if (($diagonal < $this->rows) && ($this->luMatrix[$diagonal][$diagonal] != 0.0)) { |
| 193 |
for ($row = $diagonal + 1; $row < $this->rows; ++$row) { |
| 194 |
$this->luMatrix[$row][$diagonal] /= $this->luMatrix[$diagonal][$diagonal]; |
| 195 |
} |
| 196 |
} |
| 197 |
} |
| 198 |
|
| 199 |
private function pivotB(Matrix $B) |
| 200 |
{ |
| 201 |
$X = []; |
| 202 |
foreach ($this->pivot as $rowId) { |
| 203 |
$row = $B->getRows($rowId + 1)->toArray(); |
| 204 |
$X[] = array_pop($row); |
| 205 |
} |
| 206 |
|
| 207 |
return $X; |
| 208 |
} |
| 209 |
|
| 210 |
/** |
| 211 |
* Solve A*X = B. |
| 212 |
* |
| 213 |
* @param Matrix $B a Matrix with as many rows as A and any number of columns |
| 214 |
* |
| 215 |
* @throws Exception |
| 216 |
* |
| 217 |
* @return Matrix X so that L*U*X = B(piv,:) |
| 218 |
*/ |
| 219 |
public function solve(Matrix $B) |
| 220 |
{ |
| 221 |
if ($B->rows !== $this->rows) { |
| 222 |
throw new Exception('Matrix row dimensions are not equal'); |
| 223 |
} |
| 224 |
|
| 225 |
if ($this->rows !== $this->columns) { |
| 226 |
throw new Exception('LU solve() only works on square matrices'); |
| 227 |
} |
| 228 |
|
| 229 |
if (!$this->isNonsingular()) { |
| 230 |
throw new Exception('Can only perform operation on singular matrix'); |
| 231 |
} |
| 232 |
|
| 233 |
// Copy right hand side with pivoting |
| 234 |
$nx = $B->columns; |
| 235 |
$X = $this->pivotB($B); |
| 236 |
|
| 237 |
// Solve L*Y = B(piv,:) |
| 238 |
for ($k = 0; $k < $this->columns; ++$k) { |
| 239 |
for ($i = $k + 1; $i < $this->columns; ++$i) { |
| 240 |
for ($j = 0; $j < $nx; ++$j) { |
| 241 |
$X[$i][$j] -= $X[$k][$j] * $this->luMatrix[$i][$k]; |
| 242 |
} |
| 243 |
} |
| 244 |
} |
| 245 |
|
| 246 |
// Solve U*X = Y; |
| 247 |
for ($k = $this->columns - 1; $k >= 0; --$k) { |
| 248 |
for ($j = 0; $j < $nx; ++$j) { |
| 249 |
$X[$k][$j] /= $this->luMatrix[$k][$k]; |
| 250 |
} |
| 251 |
for ($i = 0; $i < $k; ++$i) { |
| 252 |
for ($j = 0; $j < $nx; ++$j) { |
| 253 |
$X[$i][$j] -= $X[$k][$j] * $this->luMatrix[$i][$k]; |
| 254 |
} |
| 255 |
} |
| 256 |
} |
| 257 |
|
| 258 |
return new Matrix($X); |
| 259 |
} |
| 260 |
} |
| 261 |
|