请教有关“数值代数中Arnoldi 算法的块状化的比较”的问题

Meng1985

数值代数中Arnoldi 算法的块状化的比较；
%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%使用 double 数据类型%%
function [v,H] = blockArnoldi(A,m,r)
% % Input: A -- an n by n matrix
% m -- a positive integer
% r -- an Nxp block vector (v .ne. 0 assumed)
% % Output: V -- an n by m*p orthogonal matrix
% H -- a m*p by (m-1)*p upper Hessenberg matrix
if isnumeric(A)
  A = @(x)(A*x);
end
n = length(r);
p = size(r,2);
H (m*p,(m-1)*p)= 0;
v (n, m*p) = 0;
[v(:,1:p),H(1:p,1:p)] = sqr(r);
k = 0;
while k < m
     k = k +1;
     kp = k*p;
     tic;
     w = A(v(:,kp-p+1:kp));  
     toc,
for j =1:k
     jp = j*p;
     h = v(:,jp-p+1:jp)' * w;  
     w = w - v(:,jp-p+1:jp)*h;
     H(jp-p+1:jp,kp-p+1:kp) = h;   
end
    [v(:,kp+1:kp+p) ,H(kp+1:kp+p,kp-p+1:kp)] = sqr(w);  
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%% 使用 cell 数据类型

function [v,H] = blockArnoldi_cell(A,m,r)
% % Input: A -- an n by n matrix
% m -- a positive integer
% r -- an Nxp block vector (v .ne. 0 assumed)
% % Output: V -- an n by m*p orthogonal matrix
% H -- a m*p by (m-1)*p upper Hessenberg matrix
if isnumeric(A)
  A = @(x)(A*x);
end
%n = length(r);
p = size(r,2);
%H (m*p,(m-1)*p)= 0;
%v (n, m*p) = 0;
v = {}; h ={};
[v{1},h{1,1}] = sqr(r);
k = 0;
while k < m
     k = k + 1;
     w = A(v{k});   
for j =1:k
     h{j,k} = v{j}' * w;  
     w = w - v{j} * h{j,k};  
end
    [v{k+1},h{k+1,k}] = sqr(w);
end
H = zeros(p,p);
for j = 1: m-1
    for  k = j+2 : m+1
       h{k,j} = H;
    end
end
%  h
cell2mat(v);
H =cell2mat(h);
%%%%%%%%%%%%%%%%%%%%
实验结果： cell 数据类型的程序比double类型的程序快！
问题1，不是double类型数组访问是效率比cell数据类型的数组访问效率快很多吗？ 为何这里的cell 数据类型的程序比double类型的程序快？
问题2， 如何把下列程序向量化？或者有更好的方法使得下列程序执行效率快一些？
for j = 1: m-1
    for  k = j+2 : m+1
       h{k,j} = H;
    end
end
%%%%%%%%%%%
在这里先谢谢各位了！！

Meng1985

怎么没有人回帖啊.....................