An efficient method to set up a Lanczos based preconditioner for discrete ill-posed problems
|
|
- Sharleen Horton
- 5 years ago
- Views:
Transcription
1 Appled Mathematcal Appled Mathematcal Modellng (214) 1 16 Modellng An effcent method to set up a Lanczos based precondtoner for dscrete ll-posed problems Shervan Erfan a, Al Tavakol a,, Davod Khojasteh Salkuyeh b a Mathematcs Department, Val-e-Asr Unversty of Rafsanjan, Iran. b Faculty of Mathematcal Scences, Unversty of Gulan, P.O. Box 1914, Rasht, Iran Abstract Rezgh and Hossen [M. Rezgh and S. M. Hossen, Computng 88(21), 79-96] presented a Lanczos based precondtoner for dscrete ll-posed problems. Ther precondtoner s constructed by usng few steps (e.g., k) of the Lanczos bdagonalzaton and correspondng computed sngular values and rght Lanczos vectors. In ths artcle, we propose an effcent method to set up such precondtoner. Some numercal examples are gven to show the effectveness of the method. c 211 Publshed by Elsever Ltd. Keywords: Lanczos bdagonalzaton, Ill-posed problems, Precontoner, Regularzaton, Sngular value decomposton. 2 MSC: 35J65, 58J9 1. Introducton We consder large scale dscrete ll-posed problems,.e., lnear systems of equatons of the from Ax = b, A R n n, x, b R n. (1) The coeffcent matrx A s of full-rank and typcally ll-condtoned, the rght-hand sde vector b s perturbed by an error such that b = ˆb + e, whch may occur by measurement or dscretzaton errors. Here e and ˆb denote the Gaussan nose and unknown error-free rght-hand sde vectors, respectvely. These problems typcally arse, for nstance, n dscretzng lnear ll-posed problems, such as Fredholm ntegral equaton of the frst knd wth a smooth kernel. Also n these problems the two followng crtera are satsfed: (1) The sngular values of A decay gradually to zero wth no partcular gap n the spectrum. (2) The rato between the largest and the smallest nonzero sngular values s large [7]. Correspondng author Emal addresses: sh.erfan@stu.vru.ac.r (Shervan Erfan), tavakol@mal.vru.ac.r (Al Tavakol), khojasteh@gualn.ac.r, salkuyeh@gmal.com (Davod Khojasteh Salkuyeh) URL: tavakol.vru.ac.r (Al Tavakol) 1
2 S. Erfan, A. Tavakol and D. K. Salkuyeh / Appled Mathematcal Modellng (214) We would lke to obtan an approxmate soluton ˆx of the lnear system Aˆx = ˆb wth error-free unknown rghthand sde. Snce the matrx A s severely ll-condtoned, the soluton of system (1) typcally s not a sgnfcant approxmaton of ˆx even f the nose e s small. So, we use the regularzaton methods to determne a soluton that approxmates the exact soluton ˆx. The conjugate gradent method appled to normal equatons (CGLS) s a well-known teratve method for solvng ll-posed problems. Snce, teratve methods, such as CGLS, have low rate of convergence, t s possble to speed up the convergence by a sutable precondtoner [15]. All regularzaton methods use one or more regularzaton parameters specfc to the regularzaton method that controls the amount of the stablzaton mposed on the soluton, and n most cases t s necessary to choose ths parameter from gven the problem and the gven set of data. A large regularzaton parameter makes a new wellcondtoned problem, but ts soluton may be far from the exact soluton. A small regularzaton parameter generally yelds a soluton very close to the nose-contamnated exact soluton of (1), and hence ts dstance from the nosefree soluton also can be large. Thus, one can choose a regularzaton parameter to balance the error due to nose wth the error due to regularzaton. A good choce of regularzaton parameter s clearly crucal to determne useful approxmate soluton for ll-posed problems [12]. In [15], Rezgh and Hossen proposed a new precondtoner for dscrete ll-posed problems whch s produced by a few steps of Lanczos bdagonalzaton process. The computaton of ther precondtoner depends on the large sngular values of A. In ths paper, we propose a new and effcent numercal method to estmate the large sngular values of the matrx A and construct the new form of Lanczos based precondtoner, whch can be obtaned n a very small number of operatons. The paper s organzed as follows. In Secton 2, we study Fredholm ntegral equaton of the frst knd and ntroduce the quadrature rule and Galerkn method for dscretzaton of these equatons. Also, we study the treatment of the sngular values on the matrx resulted by dscretzaton. In Secton 3, we defne Lanczos bdagonalzaton for constructng Least squares va Lanczos bdagonalzaton (LSQR) method and precondtoner to approxmate the soluton of the ll-posed problems. In Secton 4, we ntroduce a new method for constructng the new form of Lanczos based precondtoner and fndng regularzaton parameter based on large sngular values of the sngular spectrum. Fnally, n Secton 5 some numercal experments are gven. 2. Fredholm Integral Equaton of the Frst Knd 2.1. The Sngular Value Expanson A classcal example n lnear ll-posed problems s a Fredholm ntegral equaton of the frst knd wth a square ntegrable kernel b a K(s, t)f(t)dt = g(s), c s d, (2) where the rght-hand sde g and kernel K are known functons and f s an unknown soluton. Fredholm equaton of the frst knd can be classfed nto two categores. The frst one arses f the kernel functon K(s, t) s smooth. In ths case, Fredholm equaton s very often extremely ll-condtoned,.e., small changes n the data cause huge changes n the unknown functon f. That means the soluton f s extremely senstve to small changes n g(s) and so specal numercal technques called regularzaton methods are requred. In the second category, the kernel k(s, t) s a sngular functon [8]. The superor analytcal tool for analyss of the frst-knd Fredholm ntegral equaton (2) wth square ntegrable kernels s the sngular value expanson (SVE) of the kernel. A kernel K s called square ntegrable f the norm K 2 = b d a c K(s, t) 2 dsdt, s bounded. By means of the SVE, any square ntegrable kernel K can be wrtten as the followng nfnte sum [7]: K(s, t) = µ u (s)v (t). =1 2
3 S. Erfan, A. Tavakol and D. K. Salkuyeh / Appled Mathematcal Modellng (214) The functons u and v are sad to be the sngular functons of K. They are orthonormal wth respect to the usual nner product,.e., { 1, f = j, (u, u j ) = (v, v j ) =, f j, where (.,.) s defned by (φ, ψ) = b a φ(t)ψ(t)dt. The numbers µ s are the sngular values of K, whch are nonnegatve and can always be ordered n non-ncreasng order so that µ 1 µ 2 µ 3. And, f there exsts a postve real number α such that the sngular values satsfy µ = O( α ), then α s called the degree of ll-posedness, and problems s characterzed as mldly or moderately ll-posed f α 1 or α > 1, respectvely. On the other hand, f µ = O(e α ), then the problem s termed severely ll-posed (see [7]). By the defnton of K, we have K 2 = = b d a c =1 b d µ u (s)v (t) µ j u j (s)v j (t)dsdt a c =1 j=1 j=1 µ µ j u (s)u j (s)v (t)v j (t)dsdt. (3) So, we have K 2 = µ 2 <. =1 Snce the sequence {µ } =1 s nonnegatve and r =1 µ2 for all r s bounded, then =1 µ2 s convergent. Also, we have =1 α s convergent f α > 1 and dvergent f α 1. Ths shows that the µ 2 must decay faster than 1. In general, f K s a contnuous partal dervatves of order p then µ s approxmately O( p 1 2 ) [11, 17]. So, we conclude that f the kernel K s smoother, then the sngular values µ s decay to zero rapdly. Ths yelds an ncrease n the number of small sngular values Dscretzaton In order to solve Fredholm ntegral equaton of the frst knd we have to dscretze t. There are essentally two man classes of methods, namely, quadrature and Galerkn methods. In the quadrature method, a quadrature rule lke the mdpont rule, the trapezodal rule and the Smpson rule, s used. The dscretzed form of (2) reads n ω j K(s, t j )f(t j ) = g(s ), = 1, 2,..., n, j=1 wth the weghts {ω j } n j=1 and the nodes {t } n =1. So, Eq. (2) s approxmated by the n n lnear system of equatons Ax = b, where A s the n n matrx wth the entres a j = ω j K(s, t j ), x s the n-vector x = [f(t 1 ), f(t 2 ),..., f(t n )] T, and b s the n-vector b = [g(s 1 ), g(s 2 ),..., g(s n )] T. 3
4 S. Erfan, A. Tavakol and D. K. Salkuyeh / Appled Mathematcal Modellng (214) In the Galerkn method, one can choose two dfferent orthonormal bass functons, {ψ } =1,...,n and {ϕ j } j=1,...,n, to expand the functons g and f respectvely, by Thus, so that, g(s) = g = n g ψ (s) and f(t) = =1 g(s) = n f j ϕ j (t). j=1 n b f j K(s, t)ϕ j (t)dt, j=1 n b f j j=1 a d The system may be wrtten n the matrx form as Ax = b, where a j = b d wth b = [g 1, g 2,..., g n ] T and x = [f 1, f 2,..., f n ] T. a c c a K(s, t)ψ (s)ϕ j (t)dsdt. K(s, t)ψ (s)ϕ j (t)dsdt, 2.3. The Sngular Value Decomposton Let A R n n. Then, the SVD of A s a decomposton of the from A = UΣV T = n u v T, where U = (u 1, u 2,..., u n ) and V = (v 1, v 2,..., v n ) are untary matrces,.e., U T U = V T V = I n, and where Σ = dag(σ 1, σ 2,..., σ n ) has non-negatve dagonal entres appearng n non-ncreasng order such that =1 σ 1 σ 2 σ n. The numbers, = 1,..., n are the sngular values of A. In contnuaton, we descrbe that an ncrease n the dmensons of A wll only ncrease the number of small sngular values. To ths end, let µ 1, µ 2,..., µ k be the large sngular values of the kernel K. As we sad n the last subsecton, f K s a smooth functon, there exsts a few large sngular values µ that yeld k n. We have (see [1]): σ 1 µ 1 σ 2 µ 2... σ k µ k. On the other hand, snce µ s are fxed, then the large sngular values would be unchanged whenever the dmenson of the matrx A s ncreased. In connecton wth dscrete ll-posed problems, two characterstc features of SVD are very often found (1) The sngular value decay gradually to zero wth no partcular gap n the spectrum. An ncrease of the dmensons of A wll ncrease the number of small sngular values. (2) The left and rght sngular vectors u and v tend to have more sgn changes n ther elements as the ndex ncreases (.e., as s decreases). Both features are consequences of the fact that the SVD of A s closely related to SVE of the underlyng kernel K. In fact, n the sngular values of A denote to many cases approxmatons of the sngular values µ of K[2]. Let us consder, for example, the test problem derve2(1,1) n [8]. The kernel K of the ntegral equaton (2) s Green s functon for the second dervatve: { s(t 1), s t, K(s, t) = t(s 1), s t, 4
5 S. Erfan, A. Tavakol and D. K. Salkuyeh / Appled Mathematcal Modellng (214) µ Fgure 1. A comparson between and µ for derve test problem. and both ntegraton ntervals are [, 1]. The sngular values and functons are gven by (see for example [2]) µ = (π) 2, u (s) = ± 2 sn(πs), = 1, 2,..., v (t) = 2 sn(πt). (4) Snce the sngular values are proportonal to 2, The problem s moderately ll-posed. Fgure 1 shows that the sngular values of dscretzed matrx of the test problem derv2(1,1) wth dmenson 1. It s clear from ths fgure that the sngular values of matrx A and the correspondng µ of kernel K are almost the same. Snce A s a full rank matrx, then ts nverse s gven by A 1 = n =1 v σ 1 u T, and therefore the soluton of Ax = b s x = A 1 b = n =1 u T b v. (5) Moreover, snce there are dscretzaton errors or lnear approxmaton errors, the dscrete ll-posed problems always contan perturbatons and error components created n tenson all sngular vectors of A. So, we cannot compute a stablzed soluton. By (5), the errors n the vector b are amplfed wth coeffcent σ 1. If s close to zero then the soluton s affected by the error e n the vector b, b = ˆb + e, ˆb e. In ths case, we have x = V Σ 1 U T b = V Σ 1 U Tˆb + V Σ 1 U T e = ˆx + x e, where ˆx = V Σ 1 U Tˆb and x e = V Σ 1 U T e. If the sngular values decay gradually to zero, then the term x e overcomes the exact soluton. Therefore, Fourer coeffcents u T b correspondng to small sngular values wth lower rate tend to zero. Hence, the terms correspondng to small sngular values overshadow the soluton. We should use the regularzaton methods to subtract or elmnate the soluton correspondng to small sngular values. 5
6 3. Lanczos Bdagonalzaton S. Erfan, A. Tavakol and D. K. Salkuyeh / Appled Mathematcal Modellng (214) For a gven matrx A R n n, the Lanczos Bdagonalzaton algorthm generates two orthogonal matrces U n and V n such that U T n AV n = B n, where B n s a real lower bdagonal matrx. Ths algorthm s descrbed as follows: Algorthm 1. Lanczos Bdagonalzaton 1. Let β 1 := b 2, u 1 := b/β 1 and v := 2. For = 1 to n do 3. p := A T u β v 1 4. α := p 2 5. v := p /α 6. q := Av α u 7. β +1 := q 2 8. u +1 := q /β Endfor The vectors u and v produced by Lanczos Bdagonalzaton algorthm are called the left and rght Lanczos vectors, respectvely. Denotng U k = [u 1, u 2,..., u k ] and V k = [v 1, v 2,..., v k ], the followng relatons can be establshed: (1) b = β 1 u 1 = β 1 U k+1 e 1, (2) AV k = U k+1 B k, (3) A T U k+1 = V k Bk T + α k+1v k+1 e T k+1, (4) Vk T V k = I k, (5) Uk+1 T U k+1 = I k+1, where I j denotes the dentty matrx of order j, e s the th unt vector and Now, suppose we want to solve B k = α 1 β 2 α 2 β αk β k+1 mn b Ax 2, x S where S denotes the k-dmensonal subspace spanned by the frst k Lanczos vectors v, = 1,..., n. The soluton whch we seek s of the from x (k) = V k y (k) for some vectors y (k) R k. Let r (k) = b Ax (k) be the correspondng resdual. From the above relatons, we get Snce the columns of U k+1 are orthonormal, we have In the step k of the LSQR algorthm we are gong to solve. r (k) = β 1 u 1 AV k y (k) = U k+1 (β 1 e 1 B k y (k) ). r (k) 2 = β 1 e 1 B k y (k) 2. mn β 1 e 1 B k y (k) 2. y (k) 6
7 S. Erfan, A. Tavakol and D. K. Salkuyeh / Appled Mathematcal Modellng (214) Now, we wrte the exact SVD of the computed B k as k B k = H k ΓQ T k = h γ q T. =1 Then we obtan where k s typcally small. x (k) = β 1 V k k =1 h 1 γ q, (6) The best value of k can be computed by one of the followng four ways: (1) In [15] the parameter k s consdered as the smallest nteger for whch σ k < τσ 1, (7) where τ s the square root of the machne precson. Here, s are the sngular values of B k [15]. (2) Generalzed Cross-Valdaton (GCV) The parameter k s chosen by GCV method that mnmzes the followng functon (see [12]): ψ k = β 1e 1 B k y (k) 2 2 (n k) 2 (3) L-Curve To determne the L-curve assocated wth LSQR, estmates of r (k) and x (k) are needed for several values of k. In the method, the corner of the L-curve gves a good balance of the soluton sze and resdual sze, so the parameter k can be chosen by the corner ths curve (see [12]). (4) Dscrete Pcard Condton A standard tool for analyzng the dscrete ll-posed problems s the dscrete Pcard plot, whch s a plot of the quanttes, u T b and ut b that arse n (5). In order to derve a meanngful regularzaton soluton n a dscrete ll-posed problem, t must satsfes the dscrete pcard condton,.e the Fourer coeffcent u T b on the average should decay to zero faster than the sngular values [8]. In ths paper, we use the dscrete Pcard condton to fnd the parameter k. Fgure 2 shows plots of the frst 2 sngular values, Fourer coeffcents u T b, and coeffcents ut b soluton for perturbed problem shaw [8] wth Gaussan nose n whch e ˆb = and e ˆb of the = As we observe, one can choose k = 8 and k = 7 for shaw problem wth e = ˆb and e = ˆb 1.1 1, respectvely. We moreover see that the dfferent perturbatons contan a lttle change n the value of parameter k. In [15], a new regularzed precondtoner obtaned by k steps of Lanczos bdagonalzaton for dscrete ll-posed problems has been ntroduced. The constructon of ths precondtoner s not based on any partcular structure of the matrces. The matrx M = V k (B T k B k ) 1 V T k + (I V k V T k ) R n n, (8) clusters approxmately the large sngular values around 1 and leaves the others unchanged.the dea here s based on the fact that the precondtoner s constructed by k n steps of Lanczos bdagonalzaton. 7
8 S. Erfan, A. Tavakol and D. K. Salkuyeh / Appled Mathematcal Modellng (214) Pcard plot(a) 1 2 Pcard plot(b) 1 15 u T b u T b /σ 1 15 u T b u T b /σ k=8 1 5 k= Fgure 2. The frst 2 sngular values, Fourer coeffcents u T b, and coeffcents ut b for the shaw test problem wth two perturbatons e = ˆb (a) and e = ˆb (b ). 4. An effectve method for computng optmal Lanczos based precondtoner We know, the matrx A can be produced by dscretzaton of Fredholm ntegral equaton of the frst knd. Solvng these equatons by methods lke Galerkn or quadrature rules usually produce an ll-posed system Ax = b that yelds a huge condton number of A. The soluton of the system Ax = b converges well, f the dmensons of A are suffcently large (see for nstance [3]). However, n ths case, the sngular values of A generated by SVD are not exact, and hence make the selecton of parameter k dffcult. In the sequel, we ntroduce a crteron for choosng a sutable regularzaton parameter k. To do so, we dscretze the Fredholm ntegral equaton on a coarse grd that produces the system Ā x = b. Then the matrx Ā resulted from ths knd of dscretzaton has less dmensons than those of A (e.g, half of A) and ts sngular values can be absolutely computed more easly than the orgnal matrx. On the other hand, as we mentoned n Secton 2, an ncrease n the dmensons of A wll ncrease the number of small sngular values. Thus, the number of large sngular values (.e., k) for A and Ā are almost the same. Moreover the number k (for the matrx Ā) can be derved va the methods lsted n Secton 3. All we need s to compute the rght Lanczos vectors V k. To do ths, we frst note that such vectors are avalable for Ā. Let v = ( v1, v 2,..., v m ) T be a rght Lanczos vector correspondng to a coarse grd. We use an nterpolaton (prolongaton) operator I k as follows: I k v = w, where w 2j 1 := v j and w 2j := 1 2 ( vj + v j+1 ), (9) for j = 1, 2,..., m. In fact, the values of coarse grd ponts are mapped unchanged to the fne grd and the values of the fne grd ponts whch are not on the coarse grd, are equal to the average of the pont values n ther neghborhood of coarse grd. In ths way, we obtan a vector of the form w = (w 1, w 2,..., w 2m 1 ) T whch s correspondng to a fne grd. It should be noted that one may apply the operator I k several tmes (e.g., r) to obtan the rght Lanczos vector for the matrx A. The followng algorthm descrbes the procedure clearly: 8
9 S. Erfan, A. Tavakol and D. K. Salkuyeh / Appled Mathematcal Modellng (214) for n=2 for n= Fgure 3. A comparson between sngular values of shaw test problem for dfferent dmensons of A. Algorthm 2. Computng rght Lanczos vectors of A from v 1. Input A = [a 1,, a n ], Ā = [ā 1,, ā m ], b R n, b R m, v R m and r N. 2. For s = 1 to r do 3. m := length(v) 4. w := zeros(2m 1, 1) 5. j := 1 6. For = 1 to m do 7. w j := v 8. If m, then 9. w j+1 = (v + v +1 )/2 1. j = j Endf 12. Endfor 13. v := w 14. EndFor 15. q := If ā T b q and a T q b, then 17. v q1 = āt b q, ĀT b 18. v q1 = at q b A T b, 19. α = vq1 v q1, 2. Else 21. q := q Go to Lne EndIf 24. Return αw R 2r (m 1)+1 Remark 1: We note that 2 r (m 1) + 1 = n and so m = (n 1)2 r + 1. (1) Hence, n dscretzaton of ll-posed problems lke Fredholm equaton of the frst knd, one can choose n such that (n 1)2 r + 1 s a natural number for some small r (e.g r < 1). 9
10 S. Erfan, A. Tavakol and D. K. Salkuyeh / Appled Mathematcal Modellng (214) v 1 R 2 v 1 R 25 v 3 R 2 v 3 R 25 v 5 R 2 v 3 R v 2 R 2 v 2 R 25 v 4 R 2 v 4 R 25 2 v R 6 v 6 R Fgure 4. A comparson between rght Lanczos vectors of shaw test problem for dfferent dmensons of A. Convergence of Algorthm 2 In order to show the convergence of Algorthm 2, we take smply for = 1, 2,..., r, the vectors u and v to be the exact and approxmated rght Lanczos vectors of the dscretzed system on the -th coarse grd, respectvely. In addton, let I +1 be the lnear nterpolaton operator such that t takes vectors v +1 n the ( + 1)-th coarse grd and produces vectors v n the -th fne-grd accordng to the rule I +1 v +1 = v. Now, we assume that whch ɛ denotes the accuracy of the approxmaton. We have: u I +1u +1 ɛ, (11) v = I +1I Ir 1 r v r, = 1, 2,..., r. (12) On the other hand, because of the boundedness of the nterpolaton operator, there exsts a constant α such that for any vectors v and w n the -th grd, I +1v +1 I +1w +1 = I +1(v +1 w +1 ) α v +1 w +1, (13) holds. Then, by (11), (12) and (13), we have: Smlarly, we have: Hence, u 1 v 1 = u 1 I 1 2 v 2 u 1 I 1 2 u 2 + I 1 2 u 2 I 1 2 v 2 ɛ 1 + α 1 u 2 v 2. u v ɛ + α u +1 v +1, = 1, 2,..., r 1. u 1 v 1 ɛ 1 + α 1 (ɛ 2 + α 2 (...(ɛ r 1 + α r 1 u r v r )...)). For the coarsest level r, we assume that u r v r ɛ r. So, ɛ = ɛ 1 + α 1 (ɛ 2 + α 2 (...(ɛ r 1 + α r 1 ɛ r )...)) s an upper bound for u 1 v 1. Now, f ɛ for = 1, 2,..., r, then u 1 v 1 and convergence s guaranteed. 1
11 Remark 2: We note that S. Erfan, A. Tavakol and D. K. Salkuyeh / Appled Mathematcal Modellng (214) u 1 v 1 = u 1 I 1 2 I I r 1 r v r, (14) where v r = v. Then, by (14) we can measure the dfference between the exact rght Lanczos vector u 1 and that of approxmated vector obtaned usng v n Algorthm 2 for dfferent values of m. Algorthm 3. Left-Precondtoned CGLS M MG A T Ax = M MG A T b 1. Let r () = b Ax (), p () = s () = M MG (A T r () ), γ = s () For j = o untl convergence do 3. t (j) = M MG p (j), 4. q (j) = At (j), 5. α (j) = γj, q (j) x (j+1) = x (j) + α j t (j), 7. r (j+1) = r (j) α j q (j), 8. s (j+1) = M MG (A T r (j+1) ), 9. γ j+1 = s (j+1) 2 2, 1. β j = γj+1 γ j, 11. p (j+1) = s (j+1) + β k p (j), 12. Endfor Let V k = [ v 1,, v k ] wth v, = 1,, k, as the rght Lanczos vectors for the coarsest grd. We defne Ik r V k = [Ik r v 1,, Ik r v k], where Ik r s the prolongaton operator and t s appled r-tmes n Algorthm 2. In fact, by Algorthm 2, Ik r v = w. Now, the new form of the precondtoner can be defned as M MG = α 2 (I r k V k )(B T k B k ) 1 (I r k V k ) T + (I α 2 (I r k V k )(I r k V k ) T ). We call ths new precondtoner by subscrpt M G, because ths precondtoner s based on an dea of prolongaton and restrcton operators n Multgrd algorthms. The constructon of regularzed nverse precondtoners M and M MG need O(kn 2 ) and O(km 2 + rm) operatons, respectvely. We know, k and r are very small n comparson wth m and n. On the other hand, by (1), the value of m s much smaller than n. Therefore, computatonal cost of the new form Lanczos based precondtoner M MG for large n s more less than that of the precondtoner M. Ths precondtoner can be used as a left or rght precondtoner wth the CGLS method [14, 15]. Algorthm 3 shows the scheme of a left precondtoned CGLS method. In fact, CGLS s a semconvergent method: For some j, n the frst j teratons, the method converges to the exact soluton, and then suddenly starts to dverge and the nose begns to enter the soluton. Methods for fndng the optmal value of such j (e.g. L-curve) can be found n [7, 9]. To llustrate the effcency of the proposed method, we consder the shaw test problem for two cases of dmensons 25 and 2. Fgure 3 shows the sngular values obtaned by SVD from the dscretzed matrx of shaw problem, wth the dmensons of 2 and 25 for the matrces A and Ā, respectvely. As we observe, the value and number of large sngular values of A and Ā are almost the same. Also, Fgure 4 shows a comparson between rght Lanczos vectors of shaw problem for dfferent dmensons of A wth the modfer coeffcent α = v11 v Numercal results In ths secton, we present some numercal examples. All experments were carred out usng MATLAB and Hansen Regularzaton [8]. We nvestgate the performance of our proposed regularzaton parameter (k) for ntroduced precondtoner and LSQR method. In the followng examples, taken from [8] the contamnated vector b and the Gaussan nose e are consdered as b = ˆb + e, where e = 1 3 ω ˆb. ω, 11
12 S. Erfan, A. Tavakol and D. K. Salkuyeh / Appled Mathematcal Modellng (214) Exact Lanczos based precondtoner wth k=4 No precondtoner Fgure 5. A comparson between the exact soluton and the soluton obtaned by precondtoned CGLS method at step 2 and the selecton k = 4 of bdagonalzaton Lanczos for derve2(2,3) problem. and ω s Gaussan random vector wth mean and standard devaton 1. In all the followng examples, we consder the coarse grd wth dmensons On ths coarse grd, the parameter k s obtaned by (7). Moreover, by usng the nterpolaton operators, the ngredents of the precondtoner M,.e. V k, σ k and B k are obtaned for fne grd. Example 5.1 Consder the ntegral equaton where 1 K(s, t)f(t)dt = g(s), s 1, K(s, t) = { s(t 1), s < t, t(s 1), s t, f(t) = g(s) = { t, t < 1 2, 1 t, t 1 2, (4s 3 3s) 24, s < 1 2, ( 4s 3 +12s 2 9s+1) 24, s 1 2. Ths s a mldly ll-posed problem;.e., ts sngular values decay slowly to zero. We dscretze ths ntegral equaton, usng the functon derve2(2,3) (see [8]), to obtan the matrx A R 2 2. We apply the ntroduced methods n Secton 4 for selected sutable regularzaton parameter on ths coarse grd. In ths problem, we have only used k = 4 steps of Lanczos bdagonalzaton for constructng the precondtoner. Fgure 5 shows the soluton of Lanczos precondtoned obtaned for k = 4 steps of the b-dagonalzaton Lanczos and unprecondtoned systems after two teraton steps. Also, Fgure 6 shows that the relatve error for precondtoned lnear systems wth selecton k = 4 s better than that of the orgnal system. Example 5.2 The nverse heat equaton used here s a Volterra ntegral equaton of the frst knd wth [, 1] as ntegraton nterval. The kernel s K(s, t) = K(s t) wth k(t) = t 3 2 2kappa π exp( 1 4kappa 2 t ). 12
13 S. Erfan, A. Tavakol and D. K. Salkuyeh / Appled Mathematcal Modellng (214) No precondtoner Lanczos based precondtoner wth k= Fgure 6. Relatve errors versus teraton steps for A T Ax = A T b of derve2(2,3) problem 1.2 Exact Lanczos based precondtoner wth k=8 No precondtoner Fgure 7. A comparson between the exact soluton and the soluton obtaned by precondtoned CGLS method wth 4 teratons and the selecton k = 8 of Lanczos bdagonalzaton process for A T Ax = A T b of heat problem. Here, the parameter kappa controls the ll-condtonng of the matrx A: kappa =5 gves a well-condtoned matrx, kappa =1 gves an ll-condtoned matrx. The default s kappa=1. The ntegral equaton s dscretzed by means of smple collocaton and the mdpont rule wth n grd ponts (see [5, 6]). An exact soluton x s constructed, and then the rght-hand sde b s produced as b = Ax. We use the code heat(n,kappa) from [8] to dscretze ths ntegral to obtan the matrx A R 2 2. In ths problem, we used the regularzaton parameter k = 8. Fgure 7 compares the soluton obtaned of precondtoned system at 4 teratons wth k = 8 steps of Lanczos bdagonalzaton and wthout precondtoner. Fgure 8 shows the relatve error wth and wthout precondtoner. Example 5.3 We dscretze a Fredholm ntegral equaton of the frst knd (2) wth [ π 2, π 2 ] as both ntegraton nterval. The kernel K and soluton f are gven by K(s, t) = (cos s + cos t) 2 ( sn u u )2, u = π(sn s + sn t), f(t) = a 1 exp( c 1 (t t 1 ) 2 ) + a 2 exp( c 2 (t t 2 ) 2 ). 13
14 S. Erfan, A. Tavakol and D. K. Salkuyeh / Appled Mathematcal Modellng (214) No precondtoner Lanczos based precondtoner wth k= Fgure 8. Relatve errors versus teraton steps, for A T Ax = A T b of heat problem. 2.5 Exact Lanczos based precondtoner wth k=7 No precondtoner Fgure 9. A comparson between the exact soluton and the soluton obtaned by precondtoned CGLS method wth 3 teratons and the selecton k = 7 of bdagonalzaton Lanczos for A T Ax = A T b of shaw problem. The parameters a 1, a 2, c 1, c 2, t 1 and t 2 are constants that determne the shape of the soluton f. In ths mplementaton, we use a 1 = 2, a 2 = 1, c 1 = 6, c 2 = 2, t 1 =.8 and t 2 =.5. The kernel and the soluton are dscretzed by smple collocaton wth n ponts to produce A and x. Then the dscrete rght-hand sde s produced by b = Ax. We used the code shaw from [7] to obtan A R 2 2. In Fgure 9, we plotted the soluton obtaned by Lanczos based precondtoner wth k = 7 at step 2 and unprecondtoned systems. Fgure 1 shows, the relatve errors from teraton steps of CGLS method for precondtoned systems wth k = 7 and orgnal system. Example 5.4 Defne the functon { 1 + cos πx φ(x) = 3, x < 3,, x 3. Then kernel K, the soluton f, and the rght-hand sde g are gven by K(s, t) = φ(s t), f(t) = φ(t), g(s) = (6 s )( πx cos 3 ) + 9 π s sn 2π 3.
15 S. Erfan, A. Tavakol and D. K. Salkuyeh / Appled Mathematcal Modellng (214) No precondtoner Lanczos based precondtoner wth k= Fgure 1. Relatve errors versus teraton steps, for A T Ax = A T b of shaw problem..16 Exact LSQR method Fgure 11. A comparson between the exact soluton and the soluton obtaned by LSQR method at k = 6 for mn b Ax of phllps problem. Both ntegraton ntervals are [ 6, 6]. We dscretze ths problem, usng the functon phllps(2)[12], to obtan the matrx A R 2 2. In the example, we used (6) to compute the approxmated soluton. Fgure 11, dsplays an exact soluton and the soluton obtaned by LSQR method for k = 6 steps of Lanczos bdagonalzaton. 6. Concluson We have studed a recently proposed precondtoner for ll-posed lnear system of equatons whch s based on the Lanczos bdagonalzaton method. Then, we have presented an effcent method to produced such precondtoner. Fnally, some numercal results have been presented to show the effectveness of the proposed method. References [1] R.C. Allen, W.R. Boland, V. Faber, G.M. Wng, Sngular values and condton numbers of Galerkn matrces arsng from lnear ntegral equatons of the frst knd, J. Math. Anal. Appl. 19(1985) [2] A. Bjorck, Numercal methods for least square problems, SIAM, Phladelpha, [3] S.C. Brenner, L.R. Scott, The mathematcal theory of fnte element methods, 3rd edton, Sprnger, 28. [4] D. Calvett, G. Golub, L. Rechel, Estmaton of the L-curve va Lanczos bdagonalzaton, BIT. 39(1999) [5] A.S. Carasso, Determnng surface temperatures from nteror observatons, SIAM J. Appl. Math. 42(1982) [6] L. Elden, The numercal soluton of a non-characterstc Cauchy problem for a parabolc equaton, n: Deuflhart, P., Harer, E. (Eds.) Numercal Treatment of Inverse Problems n Dfferental and Integral Equatons, Brkhauser, Boston,
16 S. Erfan, A. Tavakol and D. K. Salkuyeh / Appled Mathematcal Modellng (214) [7] P.C. Hansen, Rank-defcent and dscrete ll-posed problems, SIAM, Phladelpha, [8] P.C. Hansen, Regularzaton tools: a MATLAB package for analysss and soluton of dscrete ll-posed problems, Numer. Alg. 6(1994) [9] P.C. Hansen, T.K. Jensen, G. Rodrguez, An adaptve prunng algorthm for the dscrete L-curve crteron, J. Comput. Appl. Math. 198(27) [1] P.C. Hansen, Computaton of the Sngular Value Expanson, Computng 4(1998) [11] F.R. de Hoog, Revew of Fredholm equatons of the frst knd, The applcaton and numercal soluton of ntegral equatons.sjthoff and Noordhoff, pp Leyden, Netherlands, 198. [12] M.E. Klmer, D.P. Oleary, Choosng regularzaton parameter n teratve methods for ll-posed problems, SIAM J. Matr. Anal. Appl. 22(21) [13] A. Kleefeld, Numercal results of lnear Freadholm ntegral equatons of the frst knd over surfaces, Unversty of Wsconsn-Mlwaukee, 27. [14] D.P. Oleary, J.A. Smmons, A bdagonalzaton-regularzaton procedure for large scale dscretzatons of ll-posed problems, SIAM J. Sc. and Stat. Comput. 2(1981) [15] M. Rezagh, S.M. Hossen, Lanczos based precondtoner for dscrete ll-posed problems. Computng. 88(21) [16] Y. Saad, Iteratve Methods for Sparse lnear Systems, PWS press, New York, [17] F. Smthes, The egenvalues and sngular values of ntegral equatons, Proc. London Math. Soc. 43(1937),
Errors for Linear Systems
Errors for Lnear Systems When we solve a lnear system Ax b we often do not know A and b exactly, but have only approxmatons  and ˆb avalable. Then the best thng we can do s to solve ˆx ˆb exactly whch
More informationChapter 5. Solution of System of Linear Equations. Module No. 6. Solution of Inconsistent and Ill Conditioned Systems
Numercal Analyss by Dr. Anta Pal Assstant Professor Department of Mathematcs Natonal Insttute of Technology Durgapur Durgapur-713209 emal: anta.bue@gmal.com 1 . Chapter 5 Soluton of System of Lnear Equatons
More informationA New Refinement of Jacobi Method for Solution of Linear System Equations AX=b
Int J Contemp Math Scences, Vol 3, 28, no 17, 819-827 A New Refnement of Jacob Method for Soluton of Lnear System Equatons AX=b F Naem Dafchah Department of Mathematcs, Faculty of Scences Unversty of Gulan,
More informationInexact Newton Methods for Inverse Eigenvalue Problems
Inexact Newton Methods for Inverse Egenvalue Problems Zheng-jan Ba Abstract In ths paper, we survey some of the latest development n usng nexact Newton-lke methods for solvng nverse egenvalue problems.
More informationLecture 12: Discrete Laplacian
Lecture 12: Dscrete Laplacan Scrbe: Tanye Lu Our goal s to come up wth a dscrete verson of Laplacan operator for trangulated surfaces, so that we can use t n practce to solve related problems We are mostly
More informationSingular Value Decomposition: Theory and Applications
Sngular Value Decomposton: Theory and Applcatons Danel Khashab Sprng 2015 Last Update: March 2, 2015 1 Introducton A = UDV where columns of U and V are orthonormal and matrx D s dagonal wth postve real
More informationNUMERICAL DIFFERENTIATION
NUMERICAL DIFFERENTIATION 1 Introducton Dfferentaton s a method to compute the rate at whch a dependent output y changes wth respect to the change n the ndependent nput x. Ths rate of change s called the
More informationLectures - Week 4 Matrix norms, Conditioning, Vector Spaces, Linear Independence, Spanning sets and Basis, Null space and Range of a Matrix
Lectures - Week 4 Matrx norms, Condtonng, Vector Spaces, Lnear Independence, Spannng sets and Bass, Null space and Range of a Matrx Matrx Norms Now we turn to assocatng a number to each matrx. We could
More informationHongyi Miao, College of Science, Nanjing Forestry University, Nanjing ,China. (Received 20 June 2013, accepted 11 March 2014) I)ϕ (k)
ISSN 1749-3889 (prnt), 1749-3897 (onlne) Internatonal Journal of Nonlnear Scence Vol.17(2014) No.2,pp.188-192 Modfed Block Jacob-Davdson Method for Solvng Large Sparse Egenproblems Hongy Mao, College of
More informationCHAPTER 5 NUMERICAL EVALUATION OF DYNAMIC RESPONSE
CHAPTER 5 NUMERICAL EVALUATION OF DYNAMIC RESPONSE Analytcal soluton s usually not possble when exctaton vares arbtrarly wth tme or f the system s nonlnear. Such problems can be solved by numercal tmesteppng
More informationNON-CENTRAL 7-POINT FORMULA IN THE METHOD OF LINES FOR PARABOLIC AND BURGERS' EQUATIONS
IJRRAS 8 (3 September 011 www.arpapress.com/volumes/vol8issue3/ijrras_8_3_08.pdf NON-CENTRAL 7-POINT FORMULA IN THE METHOD OF LINES FOR PARABOLIC AND BURGERS' EQUATIONS H.O. Bakodah Dept. of Mathematc
More informationCME 302: NUMERICAL LINEAR ALGEBRA FALL 2005/06 LECTURE 13
CME 30: NUMERICAL LINEAR ALGEBRA FALL 005/06 LECTURE 13 GENE H GOLUB 1 Iteratve Methods Very large problems (naturally sparse, from applcatons): teratve methods Structured matrces (even sometmes dense,
More informationAppendix B. The Finite Difference Scheme
140 APPENDIXES Appendx B. The Fnte Dfference Scheme In ths appendx we present numercal technques whch are used to approxmate solutons of system 3.1 3.3. A comprehensve treatment of theoretcal and mplementaton
More informationAPPENDIX A Some Linear Algebra
APPENDIX A Some Lnear Algebra The collecton of m, n matrces A.1 Matrces a 1,1,..., a 1,n A = a m,1,..., a m,n wth real elements a,j s denoted by R m,n. If n = 1 then A s called a column vector. Smlarly,
More informationA new Approach for Solving Linear Ordinary Differential Equations
, ISSN 974-57X (Onlne), ISSN 974-5718 (Prnt), Vol. ; Issue No. 1; Year 14, Copyrght 13-14 by CESER PUBLICATIONS A new Approach for Solvng Lnear Ordnary Dfferental Equatons Fawz Abdelwahd Department of
More informationLecture 21: Numerical methods for pricing American type derivatives
Lecture 21: Numercal methods for prcng Amercan type dervatves Xaoguang Wang STAT 598W Aprl 10th, 2014 (STAT 598W) Lecture 21 1 / 26 Outlne 1 Fnte Dfference Method Explct Method Penalty Method (STAT 598W)
More informationVector Norms. Chapter 7 Iterative Techniques in Matrix Algebra. Cauchy-Bunyakovsky-Schwarz Inequality for Sums. Distances. Convergence.
Vector Norms Chapter 7 Iteratve Technques n Matrx Algebra Per-Olof Persson persson@berkeley.edu Department of Mathematcs Unversty of Calforna, Berkeley Math 128B Numercal Analyss Defnton A vector norm
More informationReport on Image warping
Report on Image warpng Xuan Ne, Dec. 20, 2004 Ths document summarzed the algorthms of our mage warpng soluton for further study, and there s a detaled descrpton about the mplementaton of these algorthms.
More informationSome Comments on Accelerating Convergence of Iterative Sequences Using Direct Inversion of the Iterative Subspace (DIIS)
Some Comments on Acceleratng Convergence of Iteratve Sequences Usng Drect Inverson of the Iteratve Subspace (DIIS) C. Davd Sherrll School of Chemstry and Bochemstry Georga Insttute of Technology May 1998
More information= = = (a) Use the MATLAB command rref to solve the system. (b) Let A be the coefficient matrix and B be the right-hand side of the system.
Chapter Matlab Exercses Chapter Matlab Exercses. Consder the lnear system of Example n Secton.. x x x y z y y z (a) Use the MATLAB command rref to solve the system. (b) Let A be the coeffcent matrx and
More information8.4 COMPLEX VECTOR SPACES AND INNER PRODUCTS
SECTION 8.4 COMPLEX VECTOR SPACES AND INNER PRODUCTS 493 8.4 COMPLEX VECTOR SPACES AND INNER PRODUCTS All the vector spaces you have studed thus far n the text are real vector spaces because the scalars
More informationNorms, Condition Numbers, Eigenvalues and Eigenvectors
Norms, Condton Numbers, Egenvalues and Egenvectors 1 Norms A norm s a measure of the sze of a matrx or a vector For vectors the common norms are: N a 2 = ( x 2 1/2 the Eucldean Norm (1a b 1 = =1 N x (1b
More informationNew Method for Solving Poisson Equation. on Irregular Domains
Appled Mathematcal Scences Vol. 6 01 no. 8 369 380 New Method for Solvng Posson Equaton on Irregular Domans J. Izadan and N. Karamooz Department of Mathematcs Facult of Scences Mashhad BranchIslamc Azad
More informationρ some λ THE INVERSE POWER METHOD (or INVERSE ITERATION) , for , or (more usually) to
THE INVERSE POWER METHOD (or INVERSE ITERATION) -- applcaton of the Power method to A some fxed constant ρ (whch s called a shft), x λ ρ If the egenpars of A are { ( λ, x ) } ( ), or (more usually) to,
More informationMMA and GCMMA two methods for nonlinear optimization
MMA and GCMMA two methods for nonlnear optmzaton Krster Svanberg Optmzaton and Systems Theory, KTH, Stockholm, Sweden. krlle@math.kth.se Ths note descrbes the algorthms used n the author s 2007 mplementatons
More informationn α j x j = 0 j=1 has a nontrivial solution. Here A is the n k matrix whose jth column is the vector for all t j=0
MODULE 2 Topcs: Lnear ndependence, bass and dmenson We have seen that f n a set of vectors one vector s a lnear combnaton of the remanng vectors n the set then the span of the set s unchanged f that vector
More informationOne-sided finite-difference approximations suitable for use with Richardson extrapolation
Journal of Computatonal Physcs 219 (2006) 13 20 Short note One-sded fnte-dfference approxmatons sutable for use wth Rchardson extrapolaton Kumar Rahul, S.N. Bhattacharyya * Department of Mechancal Engneerng,
More informationMore metrics on cartesian products
More metrcs on cartesan products If (X, d ) are metrc spaces for 1 n, then n Secton II4 of the lecture notes we defned three metrcs on X whose underlyng topologes are the product topology The purpose of
More informationLinear Approximation with Regularization and Moving Least Squares
Lnear Approxmaton wth Regularzaton and Movng Least Squares Igor Grešovn May 007 Revson 4.6 (Revson : March 004). 5 4 3 0.5 3 3.5 4 Contents: Lnear Fttng...4. Weghted Least Squares n Functon Approxmaton...
More informationStructure and Drive Paul A. Jensen Copyright July 20, 2003
Structure and Drve Paul A. Jensen Copyrght July 20, 2003 A system s made up of several operatons wth flow passng between them. The structure of the system descrbes the flow paths from nputs to outputs.
More informationConvexity preserving interpolation by splines of arbitrary degree
Computer Scence Journal of Moldova, vol.18, no.1(52), 2010 Convexty preservng nterpolaton by splnes of arbtrary degree Igor Verlan Abstract In the present paper an algorthm of C 2 nterpolaton of dscrete
More informationInner Product. Euclidean Space. Orthonormal Basis. Orthogonal
Inner Product Defnton 1 () A Eucldean space s a fnte-dmensonal vector space over the reals R, wth an nner product,. Defnton 2 (Inner Product) An nner product, on a real vector space X s a symmetrc, blnear,
More information1 Matrix representations of canonical matrices
1 Matrx representatons of canoncal matrces 2-d rotaton around the orgn: ( ) cos θ sn θ R 0 = sn θ cos θ 3-d rotaton around the x-axs: R x = 1 0 0 0 cos θ sn θ 0 sn θ cos θ 3-d rotaton around the y-axs:
More informationWorkshop: Approximating energies and wave functions Quantum aspects of physical chemistry
Workshop: Approxmatng energes and wave functons Quantum aspects of physcal chemstry http://quantum.bu.edu/pltl/6/6.pdf Last updated Thursday, November 7, 25 7:9:5-5: Copyrght 25 Dan Dll (dan@bu.edu) Department
More informationLecture Notes on Linear Regression
Lecture Notes on Lnear Regresson Feng L fl@sdueducn Shandong Unversty, Chna Lnear Regresson Problem In regresson problem, we am at predct a contnuous target value gven an nput feature vector We assume
More informationMatrix Approximation via Sampling, Subspace Embedding. 1 Solving Linear Systems Using SVD
Matrx Approxmaton va Samplng, Subspace Embeddng Lecturer: Anup Rao Scrbe: Rashth Sharma, Peng Zhang 0/01/016 1 Solvng Lnear Systems Usng SVD Two applcatons of SVD have been covered so far. Today we loo
More informationOn Finite Rank Perturbation of Diagonalizable Operators
Functonal Analyss, Approxmaton and Computaton 6 (1) (2014), 49 53 Publshed by Faculty of Scences and Mathematcs, Unversty of Nš, Serba Avalable at: http://wwwpmfnacrs/faac On Fnte Rank Perturbaton of Dagonalzable
More informationFeb 14: Spatial analysis of data fields
Feb 4: Spatal analyss of data felds Mappng rregularly sampled data onto a regular grd Many analyss technques for geophyscal data requre the data be located at regular ntervals n space and/or tme. hs s
More informationON A DETERMINATION OF THE INITIAL FUNCTIONS FROM THE OBSERVED VALUES OF THE BOUNDARY FUNCTIONS FOR THE SECOND-ORDER HYPERBOLIC EQUATION
Advanced Mathematcal Models & Applcatons Vol.3, No.3, 2018, pp.215-222 ON A DETERMINATION OF THE INITIAL FUNCTIONS FROM THE OBSERVED VALUES OF THE BOUNDARY FUNCTIONS FOR THE SECOND-ORDER HYPERBOLIC EUATION
More informationAsymptotics of the Solution of a Boundary Value. Problem for One-Characteristic Differential. Equation Degenerating into a Parabolic Equation
Nonl. Analyss and Dfferental Equatons, ol., 4, no., 5 - HIKARI Ltd, www.m-har.com http://dx.do.org/.988/nade.4.456 Asymptotcs of the Soluton of a Boundary alue Problem for One-Characterstc Dfferental Equaton
More informationSTAT 309: MATHEMATICAL COMPUTATIONS I FALL 2018 LECTURE 16
STAT 39: MATHEMATICAL COMPUTATIONS I FALL 218 LECTURE 16 1 why teratve methods f we have a lnear system Ax = b where A s very, very large but s ether sparse or structured (eg, banded, Toepltz, banded plus
More informationKernel Methods and SVMs Extension
Kernel Methods and SVMs Extenson The purpose of ths document s to revew materal covered n Machne Learnng 1 Supervsed Learnng regardng support vector machnes (SVMs). Ths document also provdes a general
More informationUsing T.O.M to Estimate Parameter of distributions that have not Single Exponential Family
IOSR Journal of Mathematcs IOSR-JM) ISSN: 2278-5728. Volume 3, Issue 3 Sep-Oct. 202), PP 44-48 www.osrjournals.org Usng T.O.M to Estmate Parameter of dstrbutons that have not Sngle Exponental Famly Jubran
More informationDifference Equations
Dfference Equatons c Jan Vrbk 1 Bascs Suppose a sequence of numbers, say a 0,a 1,a,a 3,... s defned by a certan general relatonshp between, say, three consecutve values of the sequence, e.g. a + +3a +1
More informationReview of Taylor Series. Read Section 1.2
Revew of Taylor Seres Read Secton 1.2 1 Power Seres A power seres about c s an nfnte seres of the form k = 0 k a ( x c) = a + a ( x c) + a ( x c) + a ( x c) k 2 3 0 1 2 3 + In many cases, c = 0, and the
More informationOn a direct solver for linear least squares problems
ISSN 2066-6594 Ann. Acad. Rom. Sc. Ser. Math. Appl. Vol. 8, No. 2/2016 On a drect solver for lnear least squares problems Constantn Popa Abstract The Null Space (NS) algorthm s a drect solver for lnear
More informationMATH 5630: Discrete Time-Space Model Hung Phan, UMass Lowell March 1, 2018
MATH 5630: Dscrete Tme-Space Model Hung Phan, UMass Lowell March, 08 Newton s Law of Coolng Consder the coolng of a well strred coffee so that the temperature does not depend on space Newton s law of collng
More informationSalmon: Lectures on partial differential equations. Consider the general linear, second-order PDE in the form. ,x 2
Salmon: Lectures on partal dfferental equatons 5. Classfcaton of second-order equatons There are general methods for classfyng hgher-order partal dfferental equatons. One s very general (applyng even to
More informationAdditional Codes using Finite Difference Method. 1 HJB Equation for Consumption-Saving Problem Without Uncertainty
Addtonal Codes usng Fnte Dfference Method Benamn Moll 1 HJB Equaton for Consumpton-Savng Problem Wthout Uncertanty Before consderng the case wth stochastc ncome n http://www.prnceton.edu/~moll/ HACTproect/HACT_Numercal_Appendx.pdf,
More informationInductance Calculation for Conductors of Arbitrary Shape
CRYO/02/028 Aprl 5, 2002 Inductance Calculaton for Conductors of Arbtrary Shape L. Bottura Dstrbuton: Internal Summary In ths note we descrbe a method for the numercal calculaton of nductances among conductors
More information5 The Rational Canonical Form
5 The Ratonal Canoncal Form Here p s a monc rreducble factor of the mnmum polynomal m T and s not necessarly of degree one Let F p denote the feld constructed earler n the course, consstng of all matrces
More informationThe Order Relation and Trace Inequalities for. Hermitian Operators
Internatonal Mathematcal Forum, Vol 3, 08, no, 507-57 HIKARI Ltd, wwwm-hkarcom https://doorg/0988/mf088055 The Order Relaton and Trace Inequaltes for Hermtan Operators Y Huang School of Informaton Scence
More information3.1 Expectation of Functions of Several Random Variables. )' be a k-dimensional discrete or continuous random vector, with joint PMF p (, E X E X1 E X
Statstcs 1: Probablty Theory II 37 3 EPECTATION OF SEVERAL RANDOM VARIABLES As n Probablty Theory I, the nterest n most stuatons les not on the actual dstrbuton of a random vector, but rather on a number
More informationCSci 6974 and ECSE 6966 Math. Tech. for Vision, Graphics and Robotics Lecture 21, April 17, 2006 Estimating A Plane Homography
CSc 6974 and ECSE 6966 Math. Tech. for Vson, Graphcs and Robotcs Lecture 21, Aprl 17, 2006 Estmatng A Plane Homography Overvew We contnue wth a dscusson of the major ssues, usng estmaton of plane projectve
More informationNumerical Heat and Mass Transfer
Master degree n Mechancal Engneerng Numercal Heat and Mass Transfer 06-Fnte-Dfference Method (One-dmensonal, steady state heat conducton) Fausto Arpno f.arpno@uncas.t Introducton Why we use models and
More informationLecture 3. Ax x i a i. i i
18.409 The Behavor of Algorthms n Practce 2/14/2 Lecturer: Dan Spelman Lecture 3 Scrbe: Arvnd Sankar 1 Largest sngular value In order to bound the condton number, we need an upper bound on the largest
More informationApplication of B-Spline to Numerical Solution of a System of Singularly Perturbed Problems
Mathematca Aeterna, Vol. 1, 011, no. 06, 405 415 Applcaton of B-Splne to Numercal Soluton of a System of Sngularly Perturbed Problems Yogesh Gupta Department of Mathematcs Unted College of Engneerng &
More information4DVAR, according to the name, is a four-dimensional variational method.
4D-Varatonal Data Assmlaton (4D-Var) 4DVAR, accordng to the name, s a four-dmensonal varatonal method. 4D-Var s actually a drect generalzaton of 3D-Var to handle observatons that are dstrbuted n tme. The
More informationFormulas for the Determinant
page 224 224 CHAPTER 3 Determnants e t te t e 2t 38 A = e t 2te t e 2t e t te t 2e 2t 39 If 123 A = 345, 456 compute the matrx product A adj(a) What can you conclude about det(a)? For Problems 40 43, use
More informationDeriving the X-Z Identity from Auxiliary Space Method
Dervng the X-Z Identty from Auxlary Space Method Long Chen Department of Mathematcs, Unversty of Calforna at Irvne, Irvne, CA 92697 chenlong@math.uc.edu 1 Iteratve Methods In ths paper we dscuss teratve
More informationLINEAR REGRESSION ANALYSIS. MODULE IX Lecture Multicollinearity
LINEAR REGRESSION ANALYSIS MODULE IX Lecture - 30 Multcollnearty Dr. Shalabh Department of Mathematcs and Statstcs Indan Insttute of Technology Kanpur 2 Remedes for multcollnearty Varous technques have
More informationYong Joon Ryang. 1. Introduction Consider the multicommodity transportation problem with convex quadratic cost function. 1 2 (x x0 ) T Q(x x 0 )
Kangweon-Kyungk Math. Jour. 4 1996), No. 1, pp. 7 16 AN ITERATIVE ROW-ACTION METHOD FOR MULTICOMMODITY TRANSPORTATION PROBLEMS Yong Joon Ryang Abstract. The optmzaton problems wth quadratc constrants often
More informationU.C. Berkeley CS294: Spectral Methods and Expanders Handout 8 Luca Trevisan February 17, 2016
U.C. Berkeley CS94: Spectral Methods and Expanders Handout 8 Luca Trevsan February 7, 06 Lecture 8: Spectral Algorthms Wrap-up In whch we talk about even more generalzatons of Cheeger s nequaltes, and
More informationFixed point method and its improvement for the system of Volterra-Fredholm integral equations of the second kind
MATEMATIKA, 217, Volume 33, Number 2, 191 26 c Penerbt UTM Press. All rghts reserved Fxed pont method and ts mprovement for the system of Volterra-Fredholm ntegral equatons of the second knd 1 Talaat I.
More informationLINEAR REGRESSION ANALYSIS. MODULE IX Lecture Multicollinearity
LINEAR REGRESSION ANALYSIS MODULE IX Lecture - 31 Multcollnearty Dr. Shalabh Department of Mathematcs and Statstcs Indan Insttute of Technology Kanpur 6. Rdge regresson The OLSE s the best lnear unbased
More informationThe Minimum Universal Cost Flow in an Infeasible Flow Network
Journal of Scences, Islamc Republc of Iran 17(2): 175-180 (2006) Unversty of Tehran, ISSN 1016-1104 http://jscencesutacr The Mnmum Unversal Cost Flow n an Infeasble Flow Network H Saleh Fathabad * M Bagheran
More informationFinite Element Modelling of truss/cable structures
Pet Schreurs Endhoven Unversty of echnology Department of Mechancal Engneerng Materals echnology November 3, 214 Fnte Element Modellng of truss/cable structures 1 Fnte Element Analyss of prestressed structures
More informationModule 3 LOSSY IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur
Module 3 LOSSY IMAGE COMPRESSION SYSTEMS Verson ECE IIT, Kharagpur Lesson 6 Theory of Quantzaton Verson ECE IIT, Kharagpur Instructonal Objectves At the end of ths lesson, the students should be able to:
More informationRelaxation Methods for Iterative Solution to Linear Systems of Equations
Relaxaton Methods for Iteratve Soluton to Lnear Systems of Equatons Gerald Recktenwald Portland State Unversty Mechancal Engneerng Department gerry@pdx.edu Overvew Techncal topcs Basc Concepts Statonary
More informationA Hybrid Variational Iteration Method for Blasius Equation
Avalable at http://pvamu.edu/aam Appl. Appl. Math. ISSN: 1932-9466 Vol. 10, Issue 1 (June 2015), pp. 223-229 Applcatons and Appled Mathematcs: An Internatonal Journal (AAM) A Hybrd Varatonal Iteraton Method
More informationLecture 5.8 Flux Vector Splitting
Lecture 5.8 Flux Vector Splttng 1 Flux Vector Splttng The vector E n (5.7.) can be rewrtten as E = AU (5.8.1) (wth A as gven n (5.7.4) or (5.7.6) ) whenever, the equaton of state s of the separable form
More informationComposite Hypotheses testing
Composte ypotheses testng In many hypothess testng problems there are many possble dstrbutons that can occur under each of the hypotheses. The output of the source s a set of parameters (ponts n a parameter
More informationChapter 3 Differentiation and Integration
MEE07 Computer Modelng Technques n Engneerng Chapter Derentaton and Integraton Reerence: An Introducton to Numercal Computatons, nd edton, S. yakowtz and F. zdarovsky, Mawell/Macmllan, 990. Derentaton
More information2.3 Nilpotent endomorphisms
s a block dagonal matrx, wth A Mat dm U (C) In fact, we can assume that B = B 1 B k, wth B an ordered bass of U, and that A = [f U ] B, where f U : U U s the restrcton of f to U 40 23 Nlpotent endomorphsms
More informationSolving Fractional Nonlinear Fredholm Integro-differential Equations via Hybrid of Rationalized Haar Functions
ISSN 746-7659 England UK Journal of Informaton and Computng Scence Vol. 9 No. 3 4 pp. 69-8 Solvng Fractonal Nonlnear Fredholm Integro-dfferental Equatons va Hybrd of Ratonalzed Haar Functons Yadollah Ordokhan
More informationTHE STURM-LIOUVILLE EIGENVALUE PROBLEM - A NUMERICAL SOLUTION USING THE CONTROL VOLUME METHOD
Journal of Appled Mathematcs and Computatonal Mechancs 06, 5(), 7-36 www.amcm.pcz.pl p-iss 99-9965 DOI: 0.75/jamcm.06..4 e-iss 353-0588 THE STURM-LIOUVILLE EIGEVALUE PROBLEM - A UMERICAL SOLUTIO USIG THE
More informationThe Jacobsthal and Jacobsthal-Lucas Numbers via Square Roots of Matrices
Internatonal Mathematcal Forum, Vol 11, 2016, no 11, 513-520 HIKARI Ltd, wwwm-hkarcom http://dxdoorg/1012988/mf20166442 The Jacobsthal and Jacobsthal-Lucas Numbers va Square Roots of Matrces Saadet Arslan
More informationLecture 10 Support Vector Machines II
Lecture 10 Support Vector Machnes II 22 February 2016 Taylor B. Arnold Yale Statstcs STAT 365/665 1/28 Notes: Problem 3 s posted and due ths upcomng Frday There was an early bug n the fake-test data; fxed
More informationGeneralized Linear Methods
Generalzed Lnear Methods 1 Introducton In the Ensemble Methods the general dea s that usng a combnaton of several weak learner one could make a better learner. More formally, assume that we have a set
More informationOn the Interval Zoro Symmetric Single-step Procedure for Simultaneous Finding of Polynomial Zeros
Appled Mathematcal Scences, Vol. 5, 2011, no. 75, 3693-3706 On the Interval Zoro Symmetrc Sngle-step Procedure for Smultaneous Fndng of Polynomal Zeros S. F. M. Rusl, M. Mons, M. A. Hassan and W. J. Leong
More informationU.C. Berkeley CS294: Beyond Worst-Case Analysis Luca Trevisan September 5, 2017
U.C. Berkeley CS94: Beyond Worst-Case Analyss Handout 4s Luca Trevsan September 5, 07 Summary of Lecture 4 In whch we ntroduce semdefnte programmng and apply t to Max Cut. Semdefnte Programmng Recall that
More informationarxiv: v1 [math.co] 12 Sep 2014
arxv:1409.3707v1 [math.co] 12 Sep 2014 On the bnomal sums of Horadam sequence Nazmye Ylmaz and Necat Taskara Department of Mathematcs, Scence Faculty, Selcuk Unversty, 42075, Campus, Konya, Turkey March
More informationLOW BIAS INTEGRATED PATH ESTIMATORS. James M. Calvin
Proceedngs of the 007 Wnter Smulaton Conference S G Henderson, B Bller, M-H Hseh, J Shortle, J D Tew, and R R Barton, eds LOW BIAS INTEGRATED PATH ESTIMATORS James M Calvn Department of Computer Scence
More informationGlobal Sensitivity. Tuesday 20 th February, 2018
Global Senstvty Tuesday 2 th February, 28 ) Local Senstvty Most senstvty analyses [] are based on local estmates of senstvty, typcally by expandng the response n a Taylor seres about some specfc values
More information1 GSW Iterative Techniques for y = Ax
1 for y = A I m gong to cheat here. here are a lot of teratve technques that can be used to solve the general case of a set of smultaneous equatons (wrtten n the matr form as y = A), but ths chapter sn
More informationCOMPARISON OF SOME RELIABILITY CHARACTERISTICS BETWEEN REDUNDANT SYSTEMS REQUIRING SUPPORTING UNITS FOR THEIR OPERATIONS
Avalable onlne at http://sck.org J. Math. Comput. Sc. 3 (3), No., 6-3 ISSN: 97-537 COMPARISON OF SOME RELIABILITY CHARACTERISTICS BETWEEN REDUNDANT SYSTEMS REQUIRING SUPPORTING UNITS FOR THEIR OPERATIONS
More informationProcedia Computer Science
Avalable onlne at www.scencedrect.com Proceda Proceda Computer Computer Scence Scence 1 (01) 00 (009) 589 597 000 000 Proceda Computer Scence www.elsever.com/locate/proceda Internatonal Conference on Computatonal
More informationFall 2015: Computational and Variational Methods for Inverse Problems
Fall 215: Computatonal and Varatonal Methods for Inverse Problems Georg Stadler Courant Insttute of Mathematcal Scences New York Unversty Omar Ghattas Jackson School of Geoscences Department of Mechancal
More informationCSCE 790S Background Results
CSCE 790S Background Results Stephen A. Fenner September 8, 011 Abstract These results are background to the course CSCE 790S/CSCE 790B, Quantum Computaton and Informaton (Sprng 007 and Fall 011). Each
More informationDeveloping an Improved Shift-and-Invert Arnoldi Method
Avalable at http://pvamu.edu/aam Appl. Appl. Math. ISSN: 93-9466 Vol. 5, Issue (June 00) pp. 67-80 (Prevously, Vol. 5, No. ) Applcatons and Appled Mathematcs: An Internatonal Journal (AAM) Developng an
More information10-701/ Machine Learning, Fall 2005 Homework 3
10-701/15-781 Machne Learnng, Fall 2005 Homework 3 Out: 10/20/05 Due: begnnng of the class 11/01/05 Instructons Contact questons-10701@autonlaborg for queston Problem 1 Regresson and Cross-valdaton [40
More informationComparison of the Population Variance Estimators. of 2-Parameter Exponential Distribution Based on. Multiple Criteria Decision Making Method
Appled Mathematcal Scences, Vol. 7, 0, no. 47, 07-0 HIARI Ltd, www.m-hkar.com Comparson of the Populaton Varance Estmators of -Parameter Exponental Dstrbuton Based on Multple Crtera Decson Makng Method
More informationThe lower and upper bounds on Perron root of nonnegative irreducible matrices
Journal of Computatonal Appled Mathematcs 217 (2008) 259 267 wwwelsevercom/locate/cam The lower upper bounds on Perron root of nonnegatve rreducble matrces Guang-Xn Huang a,, Feng Yn b,keguo a a College
More informationGeneral viscosity iterative method for a sequence of quasi-nonexpansive mappings
Avalable onlne at www.tjnsa.com J. Nonlnear Sc. Appl. 9 (2016), 5672 5682 Research Artcle General vscosty teratve method for a sequence of quas-nonexpansve mappngs Cuje Zhang, Ynan Wang College of Scence,
More informationA MODIFIED METHOD FOR SOLVING SYSTEM OF NONLINEAR EQUATIONS
Journal of Mathematcs and Statstcs 9 (1): 4-8, 1 ISSN 1549-644 1 Scence Publcatons do:1.844/jmssp.1.4.8 Publshed Onlne 9 (1) 1 (http://www.thescpub.com/jmss.toc) A MODIFIED METHOD FOR SOLVING SYSTEM OF
More informationThe Exact Formulation of the Inverse of the Tridiagonal Matrix for Solving the 1D Poisson Equation with the Finite Difference Method
Journal of Electromagnetc Analyss and Applcatons, 04, 6, 0-08 Publshed Onlne September 04 n ScRes. http://www.scrp.org/journal/jemaa http://dx.do.org/0.46/jemaa.04.6000 The Exact Formulaton of the Inverse
More informationFormal solvers of the RT equation
Formal solvers of the RT equaton Formal RT solvers Runge- Kutta (reference solver) Pskunov N.: 979, Master Thess Long characterstcs (Feautrer scheme) Cannon C.J.: 970, ApJ 6, 55 Short characterstcs (Hermtan
More informationThe Finite Element Method: A Short Introduction
Te Fnte Element Metod: A Sort ntroducton Wat s FEM? Te Fnte Element Metod (FEM) ntroduced by engneers n late 50 s and 60 s s a numercal tecnque for solvng problems wc are descrbed by Ordnary Dfferental
More informationSolutions Homework 4 March 5, 2018
1 Solutons Homework 4 March 5, 018 Soluton to Exercse 5.1.8: Let a IR be a translaton and c > 0 be a re-scalng. ˆb1 (cx + a) cx n + a (cx 1 + a) c x n x 1 cˆb 1 (x), whch shows ˆb 1 s locaton nvarant and
More informationPreconditioning techniques in Chebyshev collocation method for elliptic equations
Precondtonng technques n Chebyshev collocaton method for ellptc equatons Zh-We Fang Je Shen Ha-We Sun (n memory of late Professor Benyu Guo Abstract When one approxmates ellptc equatons by the spectral
More information