Hi, recently I’ve been testing the numerical stability of my algorithms. Let me just use QR decomposition as an example. I tried to test the orthogonality (relative backward error) of the computed Q following this formula: norm(Q’ * Q - I, 1)/(size(Q, 2)*eps(Float64)) That is to say, the backward…

Numerical Stability

Ralph_Smith March 6, 2020, 2:23am 2

The norm function computes a vector norm. If you scale by the full length or change to opnorm you should get more satisfactory results.

5 Likes

Topic		Replies	Views
Has anyone noticed numerical instability in Base.LinAlg.cholfact or Base.LinAlg.ishermitian? Statistics question , linearalgebra	3	879	April 4, 2018
Help with numerically unstable algorithm Numerics question , algorithm	13	874	October 20, 2023
Question regarding numerical stability among Julia versions Numerics question	16	823	April 16, 2024
Issue arising from QR decomposition changes between 0.6 to 0.7 General Usage linearalgebra	7	1098	April 7, 2019
Gram Schmidit python to julia Performance linearalgebra	10	311	September 3, 2023