Different machine precision for scalars and vectors?
1 回表示 (過去 30 日間)
古いコメントを表示
I was prearing a demonstration for my students on the topic of numerical error, and after some trial and error devised the sequence:
for i = 1:8
a = sin(i*(pi/2))
end
which reveals errors due to machine precision (a ≠ 0) when i is even. However, these errors do not show up if I store the result in a vector:
for i = 1:8
a(i) = sin(i*(pi/2))
end
Can anyone explain why? I'm guessing the reason is that somehow fewer bits are being used to encode the fractional part of the number in the latter case, but it's not clear to me why this would be the case.
1 件のコメント
David Hill
2020 年 8 月 31 日
It is the same answer.
format long
a=sin(pi);
b=sin((1:8)*pi/2);
display(b(2));
採用された回答
Jan
2020 年 8 月 31 日
The only difference is the display in the command window. Matlab tries to display vectors in an abbreviated format, execpt if you instruct it to show the full precision. David Hill suggested format long g already. Then you see, that storing the values in a vector does not change the accuracy.
0 件のコメント
その他の回答 (1 件)
the cyclist
2020 年 8 月 31 日
If you don't have a specific reason to be using a for loop here, why not also teach them canonical use of vectorized calculations?
i = 1:8;
a(i) = sin(i*(pi/2))
And either use the advice in David Hill's comment, or subtract the exact answer from a to show the error:
exact = [1 0 -1 0 1 0 -1 0];
floatingPointError = a - exact
参考
カテゴリ
Help Center および File Exchange で Ordinary Differential Equations についてさらに検索
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!