// test lup import assert from 'assert' import approx from '../../../../../tools/approx' import math from '../../../../../src/bundleAny' /** * Tests whether `Q` and `R` are the valid QR decomposition of `A`. * * Given a real matrix `A`, `Q` and `R` should be the solutions to the equation * `A = Q*R` where Q is [orthoganal](https://en.wikipedia.org/wiki/Orthogonal_matrix) and * R is [upper triangular](https://en.wikipedia.org/wiki/Triangular_matrix). * * If `A` is a complex matrix then `Q` should be a [unitary](https://en.wikipedia.org/wiki/Unitary_matrix) * * Syntax: * * math.isValidQRDecomposition(A); * * Example: * * const m = [ * [1, -1, 4], * [1, 4, -2], * [1, 4, 2], * [1, -1, 0] * ] * const result = math.qr(m) * // r = { * // Q: [ * // [0.5, -0.5, 0.5], * // [0.5, 0.5, -0.5], * // [0.5, 0.5, 0.5], * // [0.5, -0.5, -0.5], * // ], * // R: [ * // [2, 3, 2], * // [0, 5, -2], * // [0, 0, 4], * // [0, 0, 0] * // ] * // } * * isValidQRDecomposition(m, r.Q, r.R) * // true * * r.Q[2][1] = 9 * * isValidQRDecomposition(m, r.Q, r.R) * // false * * * @param {Matrix | Array} A A two dimensional matrix or array from which the QR decomposition was formed. * @param {Matrix | Array} Q A two dimensional matrix or array equal to `Q` is an QR decomposition. * @param {Matrix | Array} R A two dimensional matrix or array equal to `R` is an QR decomposition. * * @return {Boolean} Returns true if `Q` and `R` form a valid QR decomposition of `A` */ function assertValidQRDecomposition (A, Q, R) { const Asize = math.size(A).valueOf() const rows = Asize[0] const cols = Asize[1] // sizes match assert.deepStrictEqual(math.size(Q).valueOf(), [rows, rows]) assert.deepStrictEqual(math.size(R).valueOf(), [rows, cols]) // A = Q * R approx.deepEqual(math.multiply(Q, R).valueOf(), A.valueOf()) // Q has unitary (orthonormal for real A) columns // use math.equal as approx.deepEqual cannot handle complex vs real number comparision assert(math.equal(math.multiply(math.conj(math.transpose(Q)), Q).valueOf(), math.identity([Asize[0], Asize[0]]).valueOf()), 'Matrix Q is not unitary/orthonormal') // R is upper triangular for (let i = 0; i < rows; i++) { for (let j = 0; j < i && j < cols; j++) { assert(math.isZero(math.subset(R, math.index(i, j))), 'R is not an upper triangular matrix') } } // All elements on leading diagonal of R are positive for (let i = 0; i < Math.min(rows, cols); i++) { const diagonalElement = math.subset(R, math.index(i, i)) assert(!math.isNegative(math.re(diagonalElement)), 'R has elements on the leading diagonal with a negative real part (R[' + i + '][' + i + '] = ' + diagonalElement + ')') } const raw = math.qr._denseQRimpl(Array.isArray(A) ? math.matrix(A) : A) for (let i = 0; i < raw.R._data.length; ++i) { for (let j = 0; j < i && j < (raw.R._data[0] || []).length; ++j) { approx.equal(raw.R._data[i][j], 0, 1e-10) } } } describe('qr', function () { it('should decompose matrix, n x n, no permutations, array', function () { const m = [[15, 42], [20, 81]] const r = math.qr(m) // L approx.deepEqual(r.Q.valueOf(), [[0.6, -0.8], [0.8, 0.6]]) // U approx.deepEqual(r.R.valueOf(), [[25, 90], [0, 15]]) // verify assertValidQRDecomposition(m, r.Q, r.R) const m2 = [ [7.507, 9.868, 5.057], [4.482, 2.536, 9.744], [6.527, 1.094, 3.321] ] const r2 = math.qr(m2) assertValidQRDecomposition(m2, r2.Q, r2.R) }) it('should throw a helpfull error for sparse matricies', function () { const m = math.matrix([[15, 42], [20, 81]], 'sparse') assert.throws(math.qr.bind(null, m)) }) it('should decompose matrix, n x n, dense format', function () { const m = math.matrix([[15, 42], [20, 81]], 'dense') const r = math.qr(m) // Q approx.deepEqual(r.Q.valueOf(), [[0.6, -0.8], [0.8, 0.6]]) // R approx.deepEqual(r.R.valueOf(), [[25, 90], [0, 15]]) // verify assertValidQRDecomposition(m, r.Q, r.R) }) it('should decompose matrix, n x n, with a column of zeros dense format', function () { const m = math.matrix([[5, 0, 15], [223, 0, 34.5], [1, 0, 19]], 'dense') const r = math.qr(m) // Q approx.deepEqual( r.Q.valueOf(), [ [0.02241566559605479, 0.9997386855840484, -0.004483133119210979], [0.9997386855840484, -0.02243532343507404, -0.004383698101188009], [0.004483133119210979, 0.004383698101188009, 0.9999803421609812] ]) // R approx.deepEqual( r.R.valueOf(), [ [223.0582883463423, -0, 34.912399165855504], [0, -0, 14.305351889173245], [0, 0, 18.781141919779493] ]) // verify assertValidQRDecomposition(m, r.Q, r.R) }) it('should decompose matrix, m x n, m < n, dense format', function () { const m = math.matrix( [ [15, 42, -11, 9], [20, 81, 52, 112] ], 'dense' ) const r = math.qr(m) // Q approx.deepEqual( r.Q, math.matrix( [ [0.6, -0.8], [0.8, 0.6] ] )) // R approx.deepEqual( r.R, math.matrix( [ [25, 90, 35, 95], [0, 15, 40, 60] ] )) // verify assertValidQRDecomposition(m, r.Q, r.R) const m2 = math.matrix([ [7.865, 9.293, 0.534, 7.023, 9.526, 6.005, 5.007, 5.581], [3.842, 7.807, 8.208, 2.108, 3.947, 1.154, 6.086, 6.21], [3.003, 4.084, 5.593, 4.738, 9.48, 0.927, 7.294, 5.225] ]) const r2 = math.qr(m2) assertValidQRDecomposition(m2, r2.Q, r2.R) }) it('should decompose matrix, m x n, m > n, dense format', function () { const m = math.matrix( [ [8, 4], [2, -12], [9, -2], [1, 94] ], 'dense' ) const r = math.qr(m) // Q assert.deepStrictEqual( r.Q, math.matrix( [ [0.6531972647421809, -0.0050729188524001045, -0.7248169493126636, -0.21897029208715485], [0.16329931618554522, -0.13865978196560358, -0.14374377465457616, 0.9661493287513265], [0.7348469228349535, -0.07440280983520192, 0.6732450861047025, -0.034717084043718795], [0.08164965809277261, 0.9875282032672256, 0.026817368868139818, 0.13191743558805435] ] )) // R assert.deepStrictEqual( r.R, math.matrix( [ [12.24744871391589, 6.858571279792898], [0, 94.62008243496727], [0, 0], [0, 0] ] )) // verify assertValidQRDecomposition(m, r.Q, r.R) }) it('should decompose matrix, 3 x 3, zero pivote value, dense format', function () { const m = math.matrix( [ [1, 2, 3], [2, 4, 6], [4, 8, 9] ]) const r = math.qr(m) // Q approx.deepEqual( r.Q.valueOf(), [[0.21821789023599236, 0.9759000729485332, 1.1102230246251565e-16], [0.4364357804719848, -0.09759000729485323, 0.8944271909999157], [0.8728715609439696, -0.19518001458970657, -0.447213595499958] ]) // R approx.deepEqual( r.R.valueOf(), [ [4.582575694955841, 9.165151389911681, 11.129112402035613], [0, 1.9860273225978185e-15, 0.5855400437691207], [0, 0, 1.3416407864998738] ] ) // verify assertValidQRDecomposition(m, r.Q, r.R) }) it('should decompose matrix, n x n, dense format', function () { const m = math.matrix( [ [math.complex(24, 3), math.complex(10)], [math.complex(12, 53), math.complex(1.46, 10.6)], [math.complex(0.345345, 234), math.complex(1)] ]) const r = math.qr(m) // Q assert.deepStrictEqual( r.Q, math.evaluate(`[ [0.09940285751055641 + 0.012425357188819552i, 0.6771044400000075 + 0.0032268934486674216i, 0.7225638487314755 + 0.09687792016125076i], [0.049701428755278255 + 0.2195146436691456i, 0.07692808877592644 + 0.6944571280351147i, 0.00524374167953522 - 0.6790632951693036i], [0.0014303449927909801 + 0.969177860727926i, 0.009498908256891047 - 0.23073860039312136i, -0.03522342137225792 + 0.07823687113774894i] ]`)) // R assert.deepStrictEqual( r.R, math.matrix([ [math.complex(241.44175128417413, 0), math.complex(3.3948782289740067, -0.8870876675671249)], [math.complex(0, 0), math.complex(14.254103875042043, -4.440892098500626e-16)], [math.complex(0, 0), math.complex(0, 0)] ])) // verify assertValidQRDecomposition(m, r.Q, r.R) }) it('should decompose matrix, m x n, n > m, complex numbers, dense format', function () { const m = math.evaluate(`[ [-0.3264527816002377 + 2.493709974375747i, 27.144413452851555 - 95.38310595714056i, 24.851291758133694 - 31.358002980198492i, 17.60452153083572 - 58.02180107190187i, 29.062500250928192 - 57.24316264710557i, 5.699170296748263 - 65.11241969628546i, 19.819861372592023 + 25.900390198129045i, 16.557353232092076 - 37.25486567332457i], [8.548264534732331 - 47.59913064936665i, 14.40138539657334 - 90.80495969865513i, 29.343082104326758 - 15.039062252958018i, 27.20916452240602 + 25.774841219390325i, 19.38506691927698 - 95.11167912062224i, 29.17634152715012 - 95.07970712229994i, 2.1987345350210092 - 9.041770826482406i, 2.806832236244097 + 2.0385477771778966i], [24.20532702537307 + 12.879358968749457i, 25.839682426729887 - 18.102222530229938i, 29.093489513094948 - 9.581972254775465i, 12.65038940459419 - 55.38946414968438i, -0.7049513892161683 - 23.70085292748422i, 7.910814607291806 + 24.701861346839564i, 2.4219941297871004 + 28.36329723916822i, 16.535587534250833 - 38.86239252709116i], [25.78464278752434 - 59.91370905634549i, 29.424608924558413 - 19.120899022196383i, 25.6548685301034 + 6.075863297676378i, 3.693006642780766 - 63.363384338945906i, 15.716418860938354 - 73.40923022486281i, 28.9161836809681 - 58.38357844908446i, 10.13807260697836 - 3.5085542186585883i, 16.925761654754282 - 37.905623267161424i] ]`) const r = math.qr(m) // Q assert.deepStrictEqual( r.Q, math.evaluate(`[ [-0.0038074725834465403 + 0.029084550335153184i, 0.22686378024210954 - 0.8031909609489004i, -0.1539944364016218 - 0.08044026151398012i, 0.15914274660150135 - 0.4970365797781979i], [0.09969981781897692 - 0.5551565039665838i, 0.03656768230049788 - 0.4048572821234369i, 0.03460099750064215 + 0.4176688417721519i, 0.06529314053052465 + 0.5802645116992661i], [0.2823107175583003 + 0.1502140858641239i, 0.04201869101132175 - 0.25276582362981437i, 0.7610890159088707 - 0.3999596125636107i, -0.24146613640405268 + 0.18587678263056984i], [0.3007305375258921 - 0.6987834610763923i, 0.02974780206453512 + 0.2676367453654318i, 0.23430030839452232 - 0.007054866167671124i, -0.024719751847322398 - 0.5414711325141984i] ]`)) // R assert.deepStrictEqual( r.R, math.matrix([ [ math.complex(85.74002161421444, -1.7763568394002505e-15), math.complex(75.75511004703746, 4.3347264490288016), math.complex(20.511425451943854, 26.86626726613313), math.complex(27.288058950461433, -16.62801026736354), math.complex(105.22335436327181, -17.027323945468076), math.complex(109.21486260617472, 15.233872631050161), math.complex(16.361518290342467, 13.316745322711627), math.complex(28.409955756511188, -11.605326516313891) ], [ math.complex(0, 0), math.complex(121.47784233162547, 1.7763568394002505e-15), math.complex(44.01977059734889, 24.441930600590624), math.complex(38.83986358402923, 10.93198966397847), math.complex(78.56760829656308, 7.162388196994509), math.complex(72.474482997425, -8.297010771192621), math.complex(-20.270457048330027, 21.34082444731987), math.complex(33.83280850600839, 2.9469680307519037) ], [ math.complex(0, 0), math.complex(0, 0), math.complex(25.372653909655675, 5.329070518200751e-15), math.complex(46.75701662904174, -52.038112884483404), math.complex(-25.7821433027293, -35.64391269354021), math.complex(-31.014234782164266, 3.4985227007956983), math.complex(-15.936684410229294, 18.179762871924087), math.complex(33.75717971935531, -25.758933854786893) ], [ math.complex(0, 0), math.complex(0, 0), math.complex(0, 0), math.complex(69.24128415239949, 0), math.complex(14.27806840079945, 4.055317531798819), math.complex(13.583401274164364, -21.002114936285405), math.complex(-8.485891575536547, 10.384078077176659), math.complex(31.408176714183693, 17.21736552045245) ] ])) // verify assertValidQRDecomposition(m, r.Q, r.R) }) it('Prevent regression: #1669', function () { const m = math.evaluate(`[ [0, 1], [1, 0] ]`) const r = math.qr(m) // Q assert.deepStrictEqual( r.Q, math.evaluate(`[ [-0, 1], [1, 0] ]`)) // R assert.deepStrictEqual( r.R, math.evaluate(`[ [1, -0], [0, 1] ]`) ) assertValidQRDecomposition(m, r.Q, r.R) }) it('Prevent regression for complex matricies: #1669', function () { const m = math.evaluate(`[ [0, 1], [i, 0] ]`) const r = math.qr(m) // Q approx.deepEqual( r.Q, math.evaluate(`complex([ [0, 1], [i, 0] ])`) ) // R assert.deepStrictEqual( r.R, math.evaluate(`complex([ [1, -0], [0, 1] ])`) ) assertValidQRDecomposition(m, r.Q, r.R) }) })