虽然不是我写的,但我觉得很好,希望与大家分享。以下的内容转自校内: ) X7 J/ j# }* ]) }$ s* R6 {" \& S & @! s. [' L# ]8 W% `0 s' @1 N从计算数学的字面来看,应该与计算机有密切的联系,也强调 ! H) e& u( |9 E8 e; S! X" b了实践对于计算数学的重要性。也许Parlett教授的一段话能# q) O- j4 ~3 N: e8 f
最好地说明这个问题:. H- L2 n1 J8 P: c4 G* H" d
0 y: j& B! R/ k3 i0 |* K
How could someone as brilliant as von Neumann think' M/ u# O0 ^/ t6 s2 x
hard about a subject as mundane as triangular factoriz$ S( ]4 n) [% V, C
-ation of an invertible matrix and not perceive that,: K2 r$ o. b+ k& {6 n( S
with suitable pivoting, the results are impressively) S( K" c1 ~2 a7 ^9 ]/ F+ s$ z2 o
good? Partial answers can be suggested-lack of hands-on: w, N: r5 Q1 j; R, n" M% ^7 I
experience, concentration on the inverse rather than on* o! M3 x7 w% Y9 u( d B
the solution of Ax = b -but I do not find them adequate.* E. A$ j3 M" S3 N* c y5 ~* Q t
Why did Wilkinson keep the QR algorithm as a backup to a! W2 b& W' W- k
Laguerre-based method for the unsymmetric eigenproblem: [9 A. N: ]9 ~ M
for at least two years after the appearance of QR? Why: Q& P: e i9 g7 L; L# M
did more than 20 years pass before the properties of 7 M2 e0 { Z. |, r. Ythe Lanczos algorithm were understood? I believe that: ^& h4 c3 {: B. ]) _' g+ P
the explanation must involve the impediments to, \3 V- `' N# p# u" g
comprehension of the effects of finite-precision 1 h, @& R5 S' V: D garithmetic.(引自www.siam.org/siamnews/11-03/matrix.pdf)6 k7 Y* ? L( q/ B& A8 t. Q w3 L
6 M) d; q$ k3 ~, N国外的大牛有Golub,很多这个方向的大家都是他的学生。 - U, i( b( J8 s1 p- O0 d& wKahan, James Demmel, Peter Stewart, L N Trefethen,+ M& e, v, w$ ?. ^6 X2 A
Higham,这个名单可以列的很长,这些人是矩阵计算方面 6 \, S9 Q7 A' X2 i3 k+ b的大家。 ' V" ^ d- S& ]5 I. ?' O" `3 o, S- ~, A9 L% W
矩阵计算方面最经典的书应该是J H Wilkinson的《The0 y% d( h0 ]% m* w& ?
Algebraic Eigenvalue Problem》(有中译本,石钟慈等" G$ T9 \: L: o0 `$ Z
人译,《代数特征值问题》,科学出版社,学校图书馆有,; Z% n( e ]5 F, ?, ^5 w
系里有英文版的)。这本书虽然老,但是据说读一下还是 9 a, i% S* y( c2 Y很有启发的。现在的经典是Golub和 * v0 D& R8 x/ `9 K/ i9 q1 m) t$ ?, I4 jvan Loan的《Matrix Computation》(有中译本,袁亚湘译, ! `/ J+ `" n C《矩阵计算》,科学出版社),英文版的电子版可以在网上4 V0 c6 Z: X) u' f; |
找到的。其他的书有Demmel的《Applied Numerical Linear $ X0 V7 y& _! ?; TAlgebra》,Trefethen & Bau 的 ! L* ]- ^" w" B* f, ^《Numerical Linear Algebra》据说也是很好的。Yousef 4 {; F- _2 M; b6 v% Y, [# I' mSaad有两本书《Iterative methods for sparse systems》2 A, Q# V1 b7 G6 a6 O1 B4 g3 X8 ~" C6 E
和《Numerical methods for large eigenvalue problems》,' X1 i1 R2 H5 u, I! S- F7 Q& d
写的挺有意思的,在他的主页 6 O7 W3 D! L% |4 {: x(http://www-users.cs.umn.edu/~saad/) T) K% q0 g# p4 \, ~& e- S
上可以down。说到矩阵计算,还得提到Householder的一本老 2 G7 z: F8 r6 [) ^$ W% P书,《The theory of matrices in numerical analysis》7 W4 a7 c7 y! ~- Y: [
(有中译本,系里中英文版的都有)。, i+ }$ x& J, z4 |5 r+ C2 u
& K3 ?* {# E2 S! z& X8 O, ~LN Trefethen现在是剑桥大学的教授,他写的每一本书都很经典,( G$ \& j8 ]. R
前面已经到过他的几本书了,《Spectral Method in Matlab》,: h0 ~0 K( c! n- d3 }! r5 C
《Numerical Linear Algebra》,还有《Finite Difference / R8 ?9 q/ K) P2 n5 Cand Spectral methods》(在他的主页上可以9 p J6 Q. g' g5 E" A
down,http://web.comlab.ox.ac.uk/oucl/work/nick.trefethen/) " Z* v H$ I [1 H。读他的书和文章感觉也是人生的一大享受。! _1 g7 V% S4 M: y8 L3 x; @
# @8 v1 v9 l$ m6 G- F7 o" l% u4 b3 t他在Cornell大学任教时,曾上过一门课,就是阅读数值计算的经' s$ }* T5 b$ Q, ]7 }# F4 G
典文献。为此他写过一个短文,列举了数值计算中的十三篇经典文8 |2 H! x# Z- m. I
献,也许对大家有点启发。) j \6 E4 Z8 y3 O/ ?" J
& o4 e" b& F5 C. o; Z- y9 ~1. Cooley & Tukey (1965) the Fast Fourier Transform % E3 T3 d, u* L, c2. Courant, Friedrichs & Lewy (1928) finite difference methods for PDE3 z! ?2 _2 C8 S# N! W
3. Householder (1958) QR factorization of matrices$ \; Y& f) h3 U$ h
4. Curtiss & Hirschfelder (1952) stiffness of ODEs; BD formulas ! e. }& M# Y9 Y U3 ~) q$ q+ {5. de Boor (1972) calculations with B-splines 6 O4 M/ m2 t; {8 ` n6. Courant (1943) finite element methods for PDE 0 A4 v# Q, Q0 a5 k$ [7 g3 ]7. Golub & Kahan (1965) the singular value decomposition- K% \( m: u4 q9 r# W
8. Brandt (1977) multigrid algorithms ; p4 h% k& N- i! R G9. Hestenes & Stiefel (1952) the conjugate gradient iteration) i' }/ W6 V6 N4 K! J+ S
10. Fletcher & Powell (1963)optimization via quasi-Newton updates 8 I$ {( q# `5 D. [; ^: _11. Wanner, Hairer & Norsett (1978) order stars and applications to ODE4 h1 z8 H) b `' D7 w
12. Karmarkar (1984)interior pt. methods for linear prog. 7 u/ y7 E( Q9 |2 [13. Greengard & Rokhlin (1987) multipole methods for particles % }9 Y' V8 h7 ~1 s. N . } @2 k& {* \他的remark也很有意思,We were struck by how young many3 {7 D" l8 |% a2 Y8 f) J5 W1 Q2 J
of the authors were when they wrote these **s (average * t$ T# v4 G0 ^7 Xage: 34), and by how short an influential ** can be $ @& L, W# C' j(Householder: 3.3 pages, Cooley & Tukey: 4.4).这说明大家 . x. S. [; r( M) |! {; L都还是很有希望的,呵呵。 ; f6 v4 ~+ x( d' Y1 Q. j1 h" G" t3 @* |
1 \7 U9 m" s/ y8 f1 x$ K反问题无疑是计算数学中最热门的方向之一。该方向现在有如下 1 \* J6 i/ W( I$ x2 w! b) J9 b. O' C9 l2 {
几本杂志:Inverse Problems,Journal of Inverse and Ill-posed8 }' z. E8 ?# x5 `2 N; o
' k% l* V+ f) r/ ]) `- pProblems, Inverse Problems in Sciences and Engineering(以前 / X/ q: X; L" h5 s. }3 z# p- `8 v- r
叫Inverse Problems in Engineering).第一本杂志最好,第二本杂5 B5 u' [6 q' R/ d9 Y3 ^/ ?0 a3 g
& u2 ^2 _& r; d3 L
志上面有很多苏联人的工作,第三本偏向于应用。在很多高档次的 - n* j, E* z# F0 j8 a' N7 t8 e 7 f% G2 `" T/ s7 a- d+ [; s杂志中都有反问题方面的文章,比如SIAM Journal on Numerical3 c$ o. _, o" f
1 V1 B; Y7 W2 I: t* c3 ?3 m) t
Analysis,SIAM Journal on Mathematical Analysis, SIAM " o# X! u' h2 { Z. Y0 [( X+ y5 T7 J7 G( W2 M, e
Journal on Matrix Analysis and Applications,SIAM Journal on " f, s) h; ? j; G1 `& | ' O ~+ n, `7 F6 `* p3 _Scientific Computing上也有不少反问题方面的文章。 / c$ _- A. [+ F/ r/ l$ n + Q) [+ t) ^+ A在国内做反问题做的最好的应该是复旦大学的程晋老师,他在反问8 |) x- M" V2 l" T! Z
8 n; O4 t) { r5 F& q. k: _
题的理论估计方面有不少工作,南京大学的金其年老师也有不少好7 M% T' ^! o O7 j9 v