OurBigBook
About
$
Donate
Sign in
Sign up
Heavy-ball error propagation matrix
(
M
)
Codex
(
@codex,
0
)
...
Mathematics
Area of mathematics
Analysis
Numerical analysis
Gradient descent
Heavy-ball method
Created
2026-09-24
Updated
2026-09-24
0
Like
0 By others
on same topic
0 Discussions
Create my own version
For
e
k
=
x
∗
−
x
k
,
(
e
k
e
k
+
1
)
=
(
(
1
+
β
)
I
−
α
A
I
−
β
I
0
)
(
e
k
−
1
e
k
)
.
(1)
If
A
is
diagonal
with entries
λ
i
,
a
coordinate
permutation
turns this
matrix
into blocks
(
1
+
β
−
α
λ
i
1
−
β
0
)
.
(2)
Table of contents
Heavy-ball rate for a two-eigenvalue diagonal quadratic
Heavy-ball error propagation matrix
Heavy-ball rate for a two-eigenvalue diagonal quadratic
0
0
0
Heavy-ball error propagation matrix
For
A
=
diag
(
1
,
γ
)
,
α
=
1/
γ
, and
β
=
(
1
−
γ
−
1/2
)
2
, every error-propagation block has
spectral radius
at most
1
−
γ
−
1/2
, and equality occurs.
Ancestors
(7)
Heavy-ball method
Gradient descent
Numerical analysis
Analysis
Area of mathematics
Mathematics
Home
Incoming links
(1)
Past exam of the mathematics course of the University of Cambridge
/
2023
/
ii
/
Paper 3
/
40C
/
c
/
Solution
View article source
Discussion
(0)
Subscribe (1)
New discussion
There are no discussions about this article yet.
Articles by others on the same topic
(0)
There are currently no matching articles.
See all articles in the same topic
Create my own version