1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

61

62

63

64

65

66

67

68

69

70

71

72

73

74

75

76

77

78

79

80

81

82

83

84

85

86

87

88

89

90

91

92

93

94

95

96

97

98

99

100

101

102

103

104

105

106

107

108

109

110

111

112

113

114

115

116

117

118

119

120

121

122

123

124

125

126

127

128

129

130

131

132

133

134

135

136

137

138

139

140

141

142

143

144

145

146

147

148

149

150

151

152

153

154

155

156

157

158

159

160

161

162

163

164

165

166

167

168

169

170

171

172

173

174

175

176

177

178

179

180

181

182

183

184

185

186

187

188

189

190

191

192

193

194

195

196

197

198

199

200

201

202

203

204

205

206

207

208

209

210

211

212

213

214

215

216

217

218

219

220

221

222

223

224

225

226

227

228

229

230

231

232

233

234

235

236

237

238

239

240

241

242

243

244

245

246

247

248

249

250

251

252

253

254

255

256

257

258

259

260

261

262

263

264

265

266

267

268

269

270

271

272

273

274

275

276

277

278

279

280

281

282

283

284

285

286

287

288

289

290

291

292

293

294

295

296

297

298

299

300

301

302

303

304

305

306

307

308

309

310

311

312

313

314

315

316

317

318

319

320

321

322

323

324

325

326

327

328

329

330

331

332

333

334

335

336

337

338

339

340

341

342

343

344

345

346

347

348

349

350

351

352

353

354

355

356

357

358

359

360

361

362

363

364

365

366

367

368

369

370

371

372

373

374

375

376

377

378

379

380

381

382

383

384

385

386

387

388

389

390

391

392

393

394

395

396

397

398

399

400

401

402

403

404

405

406

407

408

409

410

411

412

413

414

415

416

417

418

419

420

421

422

423

424

425

426

427

428

429

430

431

432

433

434

435

436

437

438

439

440

441

442

443

444

445

446

447

448

449

450

451

452

453

454

455

456

457

458

459

460

461

462

463

464

465

466

467

468

469

470

471

472

473

474

475

476

477

478

479

480

481

482

483

484

485

486

487

488

489

490

491

492

493

494

495

496

497

498

499

500

501

502

503

504

505

506

507

508

509

510

511

512

513

514

515

516

517

518

519

520

521

522

523

524

525

526

527

528

529

530

531

532

533

534

535

536

537

538

539

540

541

542

543

544

545

546

547

548

549

550

551

552

553

554

555

556

557

558

559

560

561

562

563

564

565

566

567

568

569

570

571

572

573

574

575

576

577

578

579

580

581

582

583

584

585

586

587

588

589

590

591

592

593

594

595

596

597

598

599

600

601

602

603

604

605

606

607

608

609

610

611

612

613

614

615

616

617

618

619

620

621

622

623

624

625

626

627

628

629

630

631

632

633

634

635

636

637

638

639

640

641

642

643

644

645

646

647

648

649

650

651

652

653

654

655

656

657

658

659

660

661

662

663

664

665

666

667

668

669

670

671

672

673

674

675

676

677

678

679

680

681

682

683

684

685

686

687

688

689

690

691

692

693

694

695

696

697

698

699

700

701

702

703

704

705

706

707

708

709

710

711

712

713

714

715

716

717

718

719

720

721

722

723

724

725

726

727

728

729

730

731

732

733

734

735

736

737

738

739

740

741

742

743

744

745

746

747

748

749

750

751

752

753

754

755

756

757

758

759

760

761

762

763

764

765

766

767

768

769

770

771

772

773

774

775

776

777

778

779

780

781

782

783

784

785

786

787

788

789

790

791

792

793

794

795

796

797

798

799

800

801

802

803

804

805

806

807

808

809

810

811

812

813

814

815

816

817

818

819

820

821

822

823

824

825

826

827

828

829

830

831

832

833

834

835

836

837

838

839

840

841

842

843

844

"""Matrix equation solver routines""" 

# Author: Jeffrey Armstrong <jeff@approximatrix.com> 

# February 24, 2012 

 

# Modified: Chad Fulton <ChadFulton@gmail.com> 

# June 19, 2014 

 

# Modified: Ilhan Polat <ilhanpolat@gmail.com> 

# September 13, 2016 

 

from __future__ import division, print_function, absolute_import 

 

import warnings 

import numpy as np 

from numpy.linalg import inv, LinAlgError, norm, cond, svd 

 

from .basic import solve, solve_triangular, matrix_balance 

from .lapack import get_lapack_funcs 

from .decomp_schur import schur 

from .decomp_lu import lu 

from .decomp_qr import qr 

from ._decomp_qz import ordqz 

from .decomp import _asarray_validated 

from .special_matrices import kron, block_diag 

 

__all__ = ['solve_sylvester', 

'solve_continuous_lyapunov', 'solve_discrete_lyapunov', 

'solve_lyapunov', 

'solve_continuous_are', 'solve_discrete_are'] 

 

 

def solve_sylvester(a, b, q): 

""" 

Computes a solution (X) to the Sylvester equation :math:`AX + XB = Q`. 

 

Parameters 

---------- 

a : (M, M) array_like 

Leading matrix of the Sylvester equation 

b : (N, N) array_like 

Trailing matrix of the Sylvester equation 

q : (M, N) array_like 

Right-hand side 

 

Returns 

------- 

x : (M, N) ndarray 

The solution to the Sylvester equation. 

 

Raises 

------ 

LinAlgError 

If solution was not found 

 

Notes 

----- 

Computes a solution to the Sylvester matrix equation via the Bartels- 

Stewart algorithm. The A and B matrices first undergo Schur 

decompositions. The resulting matrices are used to construct an 

alternative Sylvester equation (``RY + YS^T = F``) where the R and S 

matrices are in quasi-triangular form (or, when R, S or F are complex, 

triangular form). The simplified equation is then solved using 

``*TRSYL`` from LAPACK directly. 

 

.. versionadded:: 0.11.0 

 

Examples 

-------- 

Given `a`, `b`, and `q` solve for `x`: 

 

>>> from scipy import linalg 

>>> a = np.array([[-3, -2, 0], [-1, -1, 3], [3, -5, -1]]) 

>>> b = np.array([[1]]) 

>>> q = np.array([[1],[2],[3]]) 

>>> x = linalg.solve_sylvester(a, b, q) 

>>> x 

array([[ 0.0625], 

[-0.5625], 

[ 0.6875]]) 

>>> np.allclose(a.dot(x) + x.dot(b), q) 

True 

 

""" 

 

# Compute the Schur decomp form of a 

r, u = schur(a, output='real') 

 

# Compute the Schur decomp of b 

s, v = schur(b.conj().transpose(), output='real') 

 

# Construct f = u'*q*v 

f = np.dot(np.dot(u.conj().transpose(), q), v) 

 

# Call the Sylvester equation solver 

trsyl, = get_lapack_funcs(('trsyl',), (r, s, f)) 

if trsyl is None: 

raise RuntimeError('LAPACK implementation does not contain a proper ' 

'Sylvester equation solver (TRSYL)') 

y, scale, info = trsyl(r, s, f, tranb='C') 

 

y = scale*y 

 

if info < 0: 

raise LinAlgError("Illegal value encountered in " 

"the %d term" % (-info,)) 

 

return np.dot(np.dot(u, y), v.conj().transpose()) 

 

 

def solve_continuous_lyapunov(a, q): 

""" 

Solves the continuous Lyapunov equation :math:`AX + XA^H = Q`. 

 

Uses the Bartels-Stewart algorithm to find :math:`X`. 

 

Parameters 

---------- 

a : array_like 

A square matrix 

 

q : array_like 

Right-hand side square matrix 

 

Returns 

------- 

x : ndarray 

Solution to the continuous Lyapunov equation 

 

See Also 

-------- 

solve_discrete_lyapunov : computes the solution to the discrete-time 

Lyapunov equation 

solve_sylvester : computes the solution to the Sylvester equation 

 

Notes 

----- 

The continuous Lyapunov equation is a special form of the Sylvester 

equation, hence this solver relies on LAPACK routine ?TRSYL. 

 

.. versionadded:: 0.11.0 

 

Examples 

-------- 

Given `a` and `q` solve for `x`: 

 

>>> from scipy import linalg 

>>> a = np.array([[-3, -2, 0], [-1, -1, 0], [0, -5, -1]]) 

>>> b = np.array([2, 4, -1]) 

>>> q = np.eye(3) 

>>> x = linalg.solve_continuous_lyapunov(a, q) 

>>> x 

array([[ -0.75 , 0.875 , -3.75 ], 

[ 0.875 , -1.375 , 5.3125], 

[ -3.75 , 5.3125, -27.0625]]) 

>>> np.allclose(a.dot(x) + x.dot(a.T), q) 

True 

""" 

 

a = np.atleast_2d(_asarray_validated(a, check_finite=True)) 

q = np.atleast_2d(_asarray_validated(q, check_finite=True)) 

 

r_or_c = float 

 

for ind, _ in enumerate((a, q)): 

if np.iscomplexobj(_): 

r_or_c = complex 

 

if not np.equal(*_.shape): 

raise ValueError("Matrix {} should be square.".format("aq"[ind])) 

 

# Shape consistency check 

if a.shape != q.shape: 

raise ValueError("Matrix a and q should have the same shape.") 

 

# Compute the Schur decomp form of a 

r, u = schur(a, output='real') 

 

# Construct f = u'*q*u 

f = u.conj().T.dot(q.dot(u)) 

 

# Call the Sylvester equation solver 

trsyl = get_lapack_funcs('trsyl', (r, f)) 

 

dtype_string = 'T' if r_or_c == float else 'C' 

y, scale, info = trsyl(r, r, f, tranb=dtype_string) 

 

if info < 0: 

raise ValueError('?TRSYL exited with the internal error ' 

'"illegal value in argument number {}.". See ' 

'LAPACK documentation for the ?TRSYL error codes.' 

''.format(-info)) 

elif info == 1: 

warnings.warn('Input "a" has an eigenvalue pair whose sum is ' 

'very close to or exactly zero. The solution is ' 

'obtained via perturbing the coefficients.', 

RuntimeWarning) 

y *= scale 

 

return u.dot(y).dot(u.conj().T) 

 

 

# For backwards compatibility, keep the old name 

solve_lyapunov = solve_continuous_lyapunov 

 

 

def _solve_discrete_lyapunov_direct(a, q): 

""" 

Solves the discrete Lyapunov equation directly. 

 

This function is called by the `solve_discrete_lyapunov` function with 

`method=direct`. It is not supposed to be called directly. 

""" 

 

lhs = kron(a, a.conj()) 

lhs = np.eye(lhs.shape[0]) - lhs 

x = solve(lhs, q.flatten()) 

 

return np.reshape(x, q.shape) 

 

 

def _solve_discrete_lyapunov_bilinear(a, q): 

""" 

Solves the discrete Lyapunov equation using a bilinear transformation. 

 

This function is called by the `solve_discrete_lyapunov` function with 

`method=bilinear`. It is not supposed to be called directly. 

""" 

eye = np.eye(a.shape[0]) 

aH = a.conj().transpose() 

aHI_inv = inv(aH + eye) 

b = np.dot(aH - eye, aHI_inv) 

c = 2*np.dot(np.dot(inv(a + eye), q), aHI_inv) 

return solve_lyapunov(b.conj().transpose(), -c) 

 

 

def solve_discrete_lyapunov(a, q, method=None): 

""" 

Solves the discrete Lyapunov equation :math:`AXA^H - X + Q = 0`. 

 

Parameters 

---------- 

a, q : (M, M) array_like 

Square matrices corresponding to A and Q in the equation 

above respectively. Must have the same shape. 

 

method : {'direct', 'bilinear'}, optional 

Type of solver. 

 

If not given, chosen to be ``direct`` if ``M`` is less than 10 and 

``bilinear`` otherwise. 

 

Returns 

------- 

x : ndarray 

Solution to the discrete Lyapunov equation 

 

See Also 

-------- 

solve_continuous_lyapunov : computes the solution to the continuous-time 

Lyapunov equation 

 

Notes 

----- 

This section describes the available solvers that can be selected by the 

'method' parameter. The default method is *direct* if ``M`` is less than 10 

and ``bilinear`` otherwise. 

 

Method *direct* uses a direct analytical solution to the discrete Lyapunov 

equation. The algorithm is given in, for example, [1]_. However it requires 

the linear solution of a system with dimension :math:`M^2` so that 

performance degrades rapidly for even moderately sized matrices. 

 

Method *bilinear* uses a bilinear transformation to convert the discrete 

Lyapunov equation to a continuous Lyapunov equation :math:`(BX+XB'=-C)` 

where :math:`B=(A-I)(A+I)^{-1}` and 

:math:`C=2(A' + I)^{-1} Q (A + I)^{-1}`. The continuous equation can be 

efficiently solved since it is a special case of a Sylvester equation. 

The transformation algorithm is from Popov (1964) as described in [2]_. 

 

.. versionadded:: 0.11.0 

 

References 

---------- 

.. [1] Hamilton, James D. Time Series Analysis, Princeton: Princeton 

University Press, 1994. 265. Print. 

http://doc1.lbfl.li/aca/FLMF037168.pdf 

.. [2] Gajic, Z., and M.T.J. Qureshi. 2008. 

Lyapunov Matrix Equation in System Stability and Control. 

Dover Books on Engineering Series. Dover Publications. 

 

Examples 

-------- 

Given `a` and `q` solve for `x`: 

 

>>> from scipy import linalg 

>>> a = np.array([[0.2, 0.5],[0.7, -0.9]]) 

>>> q = np.eye(2) 

>>> x = linalg.solve_discrete_lyapunov(a, q) 

>>> x 

array([[ 0.70872893, 1.43518822], 

[ 1.43518822, -2.4266315 ]]) 

>>> np.allclose(a.dot(x).dot(a.T)-x, -q) 

True 

 

""" 

a = np.asarray(a) 

q = np.asarray(q) 

if method is None: 

# Select automatically based on size of matrices 

if a.shape[0] >= 10: 

method = 'bilinear' 

else: 

method = 'direct' 

 

meth = method.lower() 

 

if meth == 'direct': 

x = _solve_discrete_lyapunov_direct(a, q) 

elif meth == 'bilinear': 

x = _solve_discrete_lyapunov_bilinear(a, q) 

else: 

raise ValueError('Unknown solver %s' % method) 

 

return x 

 

 

def solve_continuous_are(a, b, q, r, e=None, s=None, balanced=True): 

r""" 

Solves the continuous-time algebraic Riccati equation (CARE). 

 

The CARE is defined as 

 

.. math:: 

 

X A + A^H X - X B R^{-1} B^H X + Q = 0 

 

The limitations for a solution to exist are : 

 

* All eigenvalues of :math:`A` on the right half plane, should be 

controllable. 

 

* The associated hamiltonian pencil (See Notes), should have 

eigenvalues sufficiently away from the imaginary axis. 

 

Moreover, if ``e`` or ``s`` is not precisely ``None``, then the 

generalized version of CARE 

 

.. math:: 

 

E^HXA + A^HXE - (E^HXB + S) R^{-1} (B^HXE + S^H) + Q = 0 

 

is solved. When omitted, ``e`` is assumed to be the identity and ``s`` 

is assumed to be the zero matrix with sizes compatible with ``a`` and 

``b`` respectively. 

 

Parameters 

---------- 

a : (M, M) array_like 

Square matrix 

b : (M, N) array_like 

Input 

q : (M, M) array_like 

Input 

r : (N, N) array_like 

Nonsingular square matrix 

e : (M, M) array_like, optional 

Nonsingular square matrix 

s : (M, N) array_like, optional 

Input 

balanced : bool, optional 

The boolean that indicates whether a balancing step is performed 

on the data. The default is set to True. 

 

Returns 

------- 

x : (M, M) ndarray 

Solution to the continuous-time algebraic Riccati equation. 

 

Raises 

------ 

LinAlgError 

For cases where the stable subspace of the pencil could not be 

isolated. See Notes section and the references for details. 

 

See Also 

-------- 

solve_discrete_are : Solves the discrete-time algebraic Riccati equation 

 

Notes 

----- 

The equation is solved by forming the extended hamiltonian matrix pencil, 

as described in [1]_, :math:`H - \lambda J` given by the block matrices :: 

 

[ A 0 B ] [ E 0 0 ] 

[-Q -A^H -S ] - \lambda * [ 0 E^H 0 ] 

[ S^H B^H R ] [ 0 0 0 ] 

 

and using a QZ decomposition method. 

 

In this algorithm, the fail conditions are linked to the symmetry 

of the product :math:`U_2 U_1^{-1}` and condition number of 

:math:`U_1`. Here, :math:`U` is the 2m-by-m matrix that holds the 

eigenvectors spanning the stable subspace with 2m rows and partitioned 

into two m-row matrices. See [1]_ and [2]_ for more details. 

 

In order to improve the QZ decomposition accuracy, the pencil goes 

through a balancing step where the sum of absolute values of 

:math:`H` and :math:`J` entries (after removing the diagonal entries of 

the sum) is balanced following the recipe given in [3]_. 

 

.. versionadded:: 0.11.0 

 

References 

---------- 

.. [1] P. van Dooren , "A Generalized Eigenvalue Approach For Solving 

Riccati Equations.", SIAM Journal on Scientific and Statistical 

Computing, Vol.2(2), DOI: 10.1137/0902010 

 

.. [2] A.J. Laub, "A Schur Method for Solving Algebraic Riccati 

Equations.", Massachusetts Institute of Technology. Laboratory for 

Information and Decision Systems. LIDS-R ; 859. Available online : 

http://hdl.handle.net/1721.1/1301 

 

.. [3] P. Benner, "Symplectic Balancing of Hamiltonian Matrices", 2001, 

SIAM J. Sci. Comput., 2001, Vol.22(5), DOI: 10.1137/S1064827500367993 

 

Examples 

-------- 

Given `a`, `b`, `q`, and `r` solve for `x`: 

 

>>> from scipy import linalg 

>>> a = np.array([[4, 3], [-4.5, -3.5]]) 

>>> b = np.array([[1], [-1]]) 

>>> q = np.array([[9, 6], [6, 4.]]) 

>>> r = 1 

>>> x = linalg.solve_continuous_are(a, b, q, r) 

>>> x 

array([[ 21.72792206, 14.48528137], 

[ 14.48528137, 9.65685425]]) 

>>> np.allclose(a.T.dot(x) + x.dot(a)-x.dot(b).dot(b.T).dot(x), -q) 

True 

 

""" 

 

# Validate input arguments 

a, b, q, r, e, s, m, n, r_or_c, gen_are = _are_validate_args( 

a, b, q, r, e, s, 'care') 

 

H = np.empty((2*m+n, 2*m+n), dtype=r_or_c) 

H[:m, :m] = a 

H[:m, m:2*m] = 0. 

H[:m, 2*m:] = b 

H[m:2*m, :m] = -q 

H[m:2*m, m:2*m] = -a.conj().T 

H[m:2*m, 2*m:] = 0. if s is None else -s 

H[2*m:, :m] = 0. if s is None else s.conj().T 

H[2*m:, m:2*m] = b.conj().T 

H[2*m:, 2*m:] = r 

 

if gen_are and e is not None: 

J = block_diag(e, e.conj().T, np.zeros_like(r, dtype=r_or_c)) 

else: 

J = block_diag(np.eye(2*m), np.zeros_like(r, dtype=r_or_c)) 

 

if balanced: 

# xGEBAL does not remove the diagonals before scaling. Also 

# to avoid destroying the Symplectic structure, we follow Ref.3 

M = np.abs(H) + np.abs(J) 

M[np.diag_indices_from(M)] = 0. 

_, (sca, _) = matrix_balance(M, separate=1, permute=0) 

# do we need to bother? 

if not np.allclose(sca, np.ones_like(sca)): 

# Now impose diag(D,inv(D)) from Benner where D is 

# square root of s_i/s_(n+i) for i=0,.... 

sca = np.log2(sca) 

# NOTE: Py3 uses "Bankers Rounding: round to the nearest even" !! 

s = np.round((sca[m:2*m] - sca[:m])/2) 

sca = 2 ** np.r_[s, -s, sca[2*m:]] 

# Elementwise multiplication via broadcasting. 

elwisescale = sca[:, None] * np.reciprocal(sca) 

H *= elwisescale 

J *= elwisescale 

 

# Deflate the pencil to 2m x 2m ala Ref.1, eq.(55) 

q, r = qr(H[:, -n:]) 

H = q[:, n:].conj().T.dot(H[:, :2*m]) 

J = q[:2*m, n:].conj().T.dot(J[:2*m, :2*m]) 

 

# Decide on which output type is needed for QZ 

out_str = 'real' if r_or_c == float else 'complex' 

 

_, _, _, _, _, u = ordqz(H, J, sort='lhp', overwrite_a=True, 

overwrite_b=True, check_finite=False, 

output=out_str) 

 

# Get the relevant parts of the stable subspace basis 

if e is not None: 

u, _ = qr(np.vstack((e.dot(u[:m, :m]), u[m:, :m]))) 

u00 = u[:m, :m] 

u10 = u[m:, :m] 

 

# Solve via back-substituion after checking the condition of u00 

up, ul, uu = lu(u00) 

if 1/cond(uu) < np.spacing(1.): 

raise LinAlgError('Failed to find a finite solution.') 

 

# Exploit the triangular structure 

x = solve_triangular(ul.conj().T, 

solve_triangular(uu.conj().T, 

u10.conj().T, 

lower=True), 

unit_diagonal=True, 

).conj().T.dot(up.conj().T) 

if balanced: 

x *= sca[:m, None] * sca[:m] 

 

# Check the deviation from symmetry for lack of success 

# See proof of Thm.5 item 3 in [2] 

u_sym = u00.conj().T.dot(u10) 

n_u_sym = norm(u_sym, 1) 

u_sym = u_sym - u_sym.conj().T 

sym_threshold = np.max([np.spacing(1000.), 0.1*n_u_sym]) 

 

if norm(u_sym, 1) > sym_threshold: 

raise LinAlgError('The associated Hamiltonian pencil has eigenvalues ' 

'too close to the imaginary axis') 

 

return (x + x.conj().T)/2 

 

 

def solve_discrete_are(a, b, q, r, e=None, s=None, balanced=True): 

r""" 

Solves the discrete-time algebraic Riccati equation (DARE). 

 

The DARE is defined as 

 

.. math:: 

 

A^HXA - X - (A^HXB) (R + B^HXB)^{-1} (B^HXA) + Q = 0 

 

The limitations for a solution to exist are : 

 

* All eigenvalues of :math:`A` outside the unit disc, should be 

controllable. 

 

* The associated symplectic pencil (See Notes), should have 

eigenvalues sufficiently away from the unit circle. 

 

Moreover, if ``e`` and ``s`` are not both precisely ``None``, then the 

generalized version of DARE 

 

.. math:: 

 

A^HXA - E^HXE - (A^HXB+S) (R+B^HXB)^{-1} (B^HXA+S^H) + Q = 0 

 

is solved. When omitted, ``e`` is assumed to be the identity and ``s`` 

is assumed to be the zero matrix. 

 

Parameters 

---------- 

a : (M, M) array_like 

Square matrix 

b : (M, N) array_like 

Input 

q : (M, M) array_like 

Input 

r : (N, N) array_like 

Square matrix 

e : (M, M) array_like, optional 

Nonsingular square matrix 

s : (M, N) array_like, optional 

Input 

balanced : bool 

The boolean that indicates whether a balancing step is performed 

on the data. The default is set to True. 

 

Returns 

------- 

x : (M, M) ndarray 

Solution to the discrete algebraic Riccati equation. 

 

Raises 

------ 

LinAlgError 

For cases where the stable subspace of the pencil could not be 

isolated. See Notes section and the references for details. 

 

See Also 

-------- 

solve_continuous_are : Solves the continuous algebraic Riccati equation 

 

Notes 

----- 

The equation is solved by forming the extended symplectic matrix pencil, 

as described in [1]_, :math:`H - \lambda J` given by the block matrices :: 

 

[ A 0 B ] [ E 0 B ] 

[ -Q E^H -S ] - \lambda * [ 0 A^H 0 ] 

[ S^H 0 R ] [ 0 -B^H 0 ] 

 

and using a QZ decomposition method. 

 

In this algorithm, the fail conditions are linked to the symmetry 

of the product :math:`U_2 U_1^{-1}` and condition number of 

:math:`U_1`. Here, :math:`U` is the 2m-by-m matrix that holds the 

eigenvectors spanning the stable subspace with 2m rows and partitioned 

into two m-row matrices. See [1]_ and [2]_ for more details. 

 

In order to improve the QZ decomposition accuracy, the pencil goes 

through a balancing step where the sum of absolute values of 

:math:`H` and :math:`J` rows/cols (after removing the diagonal entries) 

is balanced following the recipe given in [3]_. If the data has small 

numerical noise, balancing may amplify their effects and some clean up 

is required. 

 

.. versionadded:: 0.11.0 

 

References 

---------- 

.. [1] P. van Dooren , "A Generalized Eigenvalue Approach For Solving 

Riccati Equations.", SIAM Journal on Scientific and Statistical 

Computing, Vol.2(2), DOI: 10.1137/0902010 

 

.. [2] A.J. Laub, "A Schur Method for Solving Algebraic Riccati 

Equations.", Massachusetts Institute of Technology. Laboratory for 

Information and Decision Systems. LIDS-R ; 859. Available online : 

http://hdl.handle.net/1721.1/1301 

 

.. [3] P. Benner, "Symplectic Balancing of Hamiltonian Matrices", 2001, 

SIAM J. Sci. Comput., 2001, Vol.22(5), DOI: 10.1137/S1064827500367993 

 

Examples 

-------- 

Given `a`, `b`, `q`, and `r` solve for `x`: 

 

>>> from scipy import linalg as la 

>>> a = np.array([[0, 1], [0, -1]]) 

>>> b = np.array([[1, 0], [2, 1]]) 

>>> q = np.array([[-4, -4], [-4, 7]]) 

>>> r = np.array([[9, 3], [3, 1]]) 

>>> x = la.solve_discrete_are(a, b, q, r) 

>>> x 

array([[-4., -4.], 

[-4., 7.]]) 

>>> R = la.solve(r + b.T.dot(x).dot(b), b.T.dot(x).dot(a)) 

>>> np.allclose(a.T.dot(x).dot(a) - x - a.T.dot(x).dot(b).dot(R), -q) 

True 

 

""" 

 

# Validate input arguments 

a, b, q, r, e, s, m, n, r_or_c, gen_are = _are_validate_args( 

a, b, q, r, e, s, 'dare') 

 

# Form the matrix pencil 

H = np.zeros((2*m+n, 2*m+n), dtype=r_or_c) 

H[:m, :m] = a 

H[:m, 2*m:] = b 

H[m:2*m, :m] = -q 

H[m:2*m, m:2*m] = np.eye(m) if e is None else e.conj().T 

H[m:2*m, 2*m:] = 0. if s is None else -s 

H[2*m:, :m] = 0. if s is None else s.conj().T 

H[2*m:, 2*m:] = r 

 

J = np.zeros_like(H, dtype=r_or_c) 

J[:m, :m] = np.eye(m) if e is None else e 

J[m:2*m, m:2*m] = a.conj().T 

J[2*m:, m:2*m] = -b.conj().T 

 

if balanced: 

# xGEBAL does not remove the diagonals before scaling. Also 

# to avoid destroying the Symplectic structure, we follow Ref.3 

M = np.abs(H) + np.abs(J) 

M[np.diag_indices_from(M)] = 0. 

_, (sca, _) = matrix_balance(M, separate=1, permute=0) 

# do we need to bother? 

if not np.allclose(sca, np.ones_like(sca)): 

# Now impose diag(D,inv(D)) from Benner where D is 

# square root of s_i/s_(n+i) for i=0,.... 

sca = np.log2(sca) 

# NOTE: Py3 uses "Bankers Rounding: round to the nearest even" !! 

s = np.round((sca[m:2*m] - sca[:m])/2) 

sca = 2 ** np.r_[s, -s, sca[2*m:]] 

# Elementwise multiplication via broadcasting. 

elwisescale = sca[:, None] * np.reciprocal(sca) 

H *= elwisescale 

J *= elwisescale 

 

# Deflate the pencil by the R column ala Ref.1 

q_of_qr, _ = qr(H[:, -n:]) 

H = q_of_qr[:, n:].conj().T.dot(H[:, :2*m]) 

J = q_of_qr[:, n:].conj().T.dot(J[:, :2*m]) 

 

# Decide on which output type is needed for QZ 

out_str = 'real' if r_or_c == float else 'complex' 

 

_, _, _, _, _, u = ordqz(H, J, sort='iuc', 

overwrite_a=True, 

overwrite_b=True, 

check_finite=False, 

output=out_str) 

 

# Get the relevant parts of the stable subspace basis 

if e is not None: 

u, _ = qr(np.vstack((e.dot(u[:m, :m]), u[m:, :m]))) 

u00 = u[:m, :m] 

u10 = u[m:, :m] 

 

# Solve via back-substituion after checking the condition of u00 

up, ul, uu = lu(u00) 

 

if 1/cond(uu) < np.spacing(1.): 

raise LinAlgError('Failed to find a finite solution.') 

 

# Exploit the triangular structure 

x = solve_triangular(ul.conj().T, 

solve_triangular(uu.conj().T, 

u10.conj().T, 

lower=True), 

unit_diagonal=True, 

).conj().T.dot(up.conj().T) 

if balanced: 

x *= sca[:m, None] * sca[:m] 

 

# Check the deviation from symmetry for lack of success 

# See proof of Thm.5 item 3 in [2] 

u_sym = u00.conj().T.dot(u10) 

n_u_sym = norm(u_sym, 1) 

u_sym = u_sym - u_sym.conj().T 

sym_threshold = np.max([np.spacing(1000.), 0.1*n_u_sym]) 

 

if norm(u_sym, 1) > sym_threshold: 

raise LinAlgError('The associated symplectic pencil has eigenvalues' 

'too close to the unit circle') 

 

return (x + x.conj().T)/2 

 

 

def _are_validate_args(a, b, q, r, e, s, eq_type='care'): 

""" 

A helper function to validate the arguments supplied to the 

Riccati equation solvers. Any discrepancy found in the input 

matrices leads to a ``ValueError`` exception. 

 

Essentially, it performs: 

 

- a check whether the input is free of NaN and Infs. 

- a pass for the data through ``numpy.atleast_2d()`` 

- squareness check of the relevant arrays, 

- shape consistency check of the arrays, 

- singularity check of the relevant arrays, 

- symmetricity check of the relevant matrices, 

- a check whether the regular or the generalized version is asked. 

 

This function is used by ``solve_continuous_are`` and 

``solve_discrete_are``. 

 

Parameters 

---------- 

a, b, q, r, e, s : array_like 

Input data 

eq_type : str 

Accepted arguments are 'care' and 'dare'. 

 

Returns 

------- 

a, b, q, r, e, s : ndarray 

Regularized input data 

m, n : int 

shape of the problem 

r_or_c : type 

Data type of the problem, returns float or complex 

gen_or_not : bool 

Type of the equation, True for generalized and False for regular ARE. 

 

""" 

 

if not eq_type.lower() in ('dare', 'care'): 

raise ValueError("Equation type unknown. " 

"Only 'care' and 'dare' is understood") 

 

a = np.atleast_2d(_asarray_validated(a, check_finite=True)) 

b = np.atleast_2d(_asarray_validated(b, check_finite=True)) 

q = np.atleast_2d(_asarray_validated(q, check_finite=True)) 

r = np.atleast_2d(_asarray_validated(r, check_finite=True)) 

 

# Get the correct data types otherwise Numpy complains 

# about pushing complex numbers into real arrays. 

r_or_c = complex if np.iscomplexobj(b) else float 

 

for ind, mat in enumerate((a, q, r)): 

if np.iscomplexobj(mat): 

r_or_c = complex 

 

if not np.equal(*mat.shape): 

raise ValueError("Matrix {} should be square.".format("aqr"[ind])) 

 

# Shape consistency checks 

m, n = b.shape 

if m != a.shape[0]: 

raise ValueError("Matrix a and b should have the same number of rows.") 

if m != q.shape[0]: 

raise ValueError("Matrix a and q should have the same shape.") 

if n != r.shape[0]: 

raise ValueError("Matrix b and r should have the same number of cols.") 

 

# Check if the data matrices q, r are (sufficiently) hermitian 

for ind, mat in enumerate((q, r)): 

if norm(mat - mat.conj().T, 1) > np.spacing(norm(mat, 1))*100: 

raise ValueError("Matrix {} should be symmetric/hermitian." 

"".format("qr"[ind])) 

 

# Continuous time ARE should have a nonsingular r matrix. 

if eq_type == 'care': 

min_sv = svd(r, compute_uv=False)[-1] 

if min_sv == 0. or min_sv < np.spacing(1.)*norm(r, 1): 

raise ValueError('Matrix r is numerically singular.') 

 

# Check if the generalized case is required with omitted arguments 

# perform late shape checking etc. 

generalized_case = e is not None or s is not None 

 

if generalized_case: 

if e is not None: 

e = np.atleast_2d(_asarray_validated(e, check_finite=True)) 

if not np.equal(*e.shape): 

raise ValueError("Matrix e should be square.") 

if m != e.shape[0]: 

raise ValueError("Matrix a and e should have the same shape.") 

# numpy.linalg.cond doesn't check for exact zeros and 

# emits a runtime warning. Hence the following manual check. 

min_sv = svd(e, compute_uv=False)[-1] 

if min_sv == 0. or min_sv < np.spacing(1.) * norm(e, 1): 

raise ValueError('Matrix e is numerically singular.') 

if np.iscomplexobj(e): 

r_or_c = complex 

if s is not None: 

s = np.atleast_2d(_asarray_validated(s, check_finite=True)) 

if s.shape != b.shape: 

raise ValueError("Matrix b and s should have the same shape.") 

if np.iscomplexobj(s): 

r_or_c = complex 

 

return a, b, q, r, e, s, m, n, r_or_c, generalized_case