Neumann series explained
. This has the generator form
where
is the
k-times repeated application of
;
is the
identity operator
and
for
. This is a special case of the generalization of a
geometric series of real or complex numbers to a geometric series of operators. The generalized initial term of the series is the identity operator
and the generalized common ratio of the series is the operator
The series is named after the mathematician Carl Neumann, who used it in 1877 in the context of potential theory. The Neumann series is used in functional analysis. It is closely connected to the resolvent formalism for studying the spectrum of bounded operators and, applied from the left to a function, it forms the Liouville-Neumann series that formally solves Fredholm integral equations.
Properties
Suppose that
is a bounded linear operator on the
normed vector space
. If the Neumann series
converges in the
operator norm, then
is
invertible and its inverse is the series:
,where
is the
identity operator in
. To see why, consider the partial sums
.Then we have
\limn(I-T)Sn=\limn
Tk-
Tk+1\right)=\limn\left(I-Tn+1\right)=I
This result on operators is analogous to
geometric series in
.
One case in which convergence is guaranteed is when
is a
Banach space and
in the operator norm; another compatible case is that
converges. However, there are also results which give weaker conditions under which the series converges.
Example
Let
be given by:
\begin{pmatrix}
0&
&
&0&
&
&0
\end{pmatrix}.
For the Neumann series
to converge to
as
goes to infinity, the spectral norm of
must be smaller than unity. This norm is about
as can be checked numerically (e.g., by computing the singular values),confirming that the Neumann series converges.
Approximate matrix inversion
A truncated Neumann series can be used for approximate matrix inversion. To approximate the inverse of an invertible matrix
, consider that
\begin{align}
A-1&=(I-I+A)-1\\
&=(I-(I-A))-1\\
&=(I-T)-1\end{align}
for
Then, using the Neumann series identity that
if the appropriate norm condition on
is satisfied,
Since these terms shrink with increasing
given the conditions on the norm, then truncating the series at some finite
may give a practical approximation to the inverse matrix:
The set of invertible operators is open
A corollary is that the set of invertible operators between two Banach spaces
and
is open in the topology induced by the operator norm. Indeed, let
be an invertible operator and let
be another operator. If
, then
is also invertible. Since
, the Neumann series
is convergent. Therefore, we have
T-1S=(I-(I-S-1T))-1=
-S-1T)k
Taking the norms, we get
The norm of
can be bounded by
|T-1|\le\tfrac{1}{1-q}|S-1| where q=|S-T||S-1|.
Applications
The Neumann series has been used for linear data detection in massive multiuser multiple-input multiple-output (MIMO) wireless systems. Using a truncated Neumann series avoids computation of an explicit matrix inverse, which reduces the complexity of linear data detection from cubic to square.[1]
Another application is the theory of propagation graphs which takes advantage of Neumann series to derive closed form expressions for transfer functions.
References
- Book: Werner, Dirk. 2005. Funktionalanalysis . de. Springer Verlag. 3-540-43586-7.
Notes and References
- Book: Wu. M.. Yin. B.. Vosoughi. A.. Studer. C.. Cavallaro. J. R.. Dick. C.. 2013 IEEE International Symposium on Circuits and Systems (ISCAS2013) . Approximate matrix inversion for high-throughput data detection in the large-scale MIMO uplink . May 2013. 2155–2158. 10.1109/ISCAS.2013.6572301. 1911/75011. 978-1-4673-5762-3 . 389966 . free.