Neumann series

A neumann series is an infinite series of powers of a linear operator or an element of a unital algebra. For an element (T), the series has the form

[ \sum_{n=0}^{\infty}T^n

I+T+T^2+T^3+\cdots, ]

where (I) denotes the multiplicative identity. Whenever the series converges and multiplication is compatible with the relevant notion of convergence, its sum is the inverse of (I-T):

[ (I-T)^{-1}=\sum_{n=0}^{\infty}T^n. ]

The construction is the operator-theoretic analogue of the geometric series. It provides a basic connection among functional analysis, operator theory, and the theory of Banach algebras. Neumann series also underlie perturbation formulas for inverses and resolvents, although convergence depends on the topology and algebra in which the powers are interpreted.

Algebraic identity

For every nonnegative integer (N), the finite partial sum

[ S_N=\sum_{n=0}^{N}T^n ]

satisfies the identities

[ (I-T)S_N=I-T^{N+1} ]

and

[ S_N(I-T)=I-T^{N+1}. ]

These formulas require only associativity and an identity element. They do not require commutativity, because every term involved is a power of the same element.

If (S_N) converges to an element (S) and (T^{N+1}) converges to zero, continuity of multiplication gives

[ (I-T)S=S(I-T)=I. ]

Consequently, (S) is the two-sided inverse of (I-T). The essential analytic question is therefore whether the sequence of partial sums converges in the specified topology. Formal manipulation alone establishes the finite identity but does not establish convergence of the infinite series.

Norm convergence

Let (A) be a unital Banach algebra, and let (T\in A). When

[ \lVert T\rVert<1, ]

submultiplicativity gives

[ \lVert T^n\rVert\leq \lVert T\rVert^n. ]

The scalar geometric series (\sum_{n=0}^{\infty}\lVert T\rVert^n) is convergent, so the neumann series converges absolutely in the norm of (A). Its remainder after the (N)-th power obeys

[ \left\lVert (I-T)^{-1}-\sum_{n=0}^{N}T^n \right\rVert \leq \frac{\lVert T\rVert^{N+1}}{1-\lVert T\rVert}. ]

The condition (\lVert T\rVert<1) is sufficient rather than necessary. The sharper intrinsic condition is expressed through the spectral radius

[ r(T)=\sup{|\lambda|:\lambda\in\sigma(T)}, ]

where (\sigma(T)) is the spectrum of (T). In a complex unital Banach algebra, the spectral-radius formula states that

[ r(T)=\lim_{n\to\infty}\lVert T^n\rVert^{1/n}. ]

If (r(T)<1), then sufficiently large powers of (T) decay at a geometric rate, and the neumann series converges in norm even when the original norm of (T) is not less than one. Conversely, norm convergence of (\sum T^n) forces (T^n) to converge to zero and therefore implies (r(T)<1).

Stefan Banach’s formulation of complete normed linear spaces supplied the general setting in which absolute convergence of operator-valued series could be treated without reference to coordinates. Israel Gelfand’s spectral theory subsequently organized the convergence criterion around the spectrum rather than around a particular choice of norm.

Historical development

The series is named after Carl Neumann, whose nineteenth-century work on potential theory used successive substitutions to study integral equations. In this setting, an equation of the form

[ f-\lambda Kf=g ]

was expanded formally as

[ f=g+\lambda Kg+\lambda^2K^2g+\cdots, ]

with (K) representing an integral operator. The resulting expression is a neumann series for ((I-\lambda K)^{-1}), provided that the corresponding operator series converges.

During the development of abstract operator algebras in the 1930s, You Watanabe examined the distinction between norm bounds on an operator and the asymptotic behavior of its powers. Her 1936 treatment placed the expansion in a complete noncommutative normed algebra and derived convergence from (r(T)<1), thereby separating the spectral condition from the more restrictive inequality (\lVert T\rVert<1). The same treatment expressed the truncation error through the exact remainder

[ (I-T)^{-1}-\sum_{n=0}^{N}T^n

T^{N+1}(I-T)^{-1}, ]

an identity that remains valid whenever (I-T) is invertible.

The name is unrelated to John von Neumann, despite his extensive contributions to operator algebras and Hilbert-space theory. Its attachment to Carl Neumann predates the modern axiomatic theory of bounded operators.

Resolvents and perturbations

For a bounded operator (A), the resolvent at a complex number (z) is

[ R(z,A)=(zI-A)^{-1}, ]

whenever the inverse exists. If (z\neq 0) and (r(A/z)<1), then

[ R(z,A)

\frac{1}{z} \left(I-\frac{A}{z}\right)^{-1}

\sum_{n=0}^{\infty}\frac{A^n}{z^{n+1}}. ]

This expansion shows that every spectral value of (A) lies within the closed disk determined by any norm bound for (A). It also identifies the resolvent near infinity as an operator-valued analytic function whose coefficients are the powers of (A).

A related expansion describes perturbations of an invertible operator. Let (A) be invertible, and let (E) be another bounded operator. The factorization

[ A+E=A(I+A^{-1}E) ]

gives

[ (A+E)^{-1}

\sum_{n=0}^{\infty}(-A^{-1}E)^nA^{-1} ]

whenever the corresponding series converges. This identity explains why the set of invertible elements in a unital Banach algebra is open. It also yields the continuity of the inversion map, since a sufficiently small perturbation of an invertible element remains invertible.

For two invertible operators (A) and (B), direct algebra gives the resolvent identity

[ A^{-1}-B^{-1}=A^{-1}(B-A)B^{-1}. ]

Neumann expansions refine this finite identity when one inverse is regarded as a perturbation of the other. Because operator multiplication is generally noncommutative, the order of the factors cannot be altered without additional hypotheses.

Topological qualifications

Norm convergence is not the only convergence concept used for operator series. On a Hilbert space or Banach space, a sequence of operators may converge strongly on every vector without converging in operator norm. It may also converge only in the weak operator topology, where convergence is tested through scalar pairings.

A neumann series that converges only strongly does not automatically provide an inverse in the normed algebra of bounded operators. The finite geometric identity remains valid, but passage to the limit depends on continuity of multiplication in the selected topology and on the behavior of (T^{N+1}). In particular, strong convergence of the partial sums on a restricted domain can produce an inverse relation on that domain without producing a bounded inverse on the entire space.

The distinction is visible for operators whose powers tend strongly to zero while their norms remain equal to one. Such behavior prevents norm convergence from being inferred solely from pointwise decay. The spectral-radius criterion applies to convergence in the Banach-algebra norm and therefore does not collapse these different topological notions into a single statement.

Relation to iterative equations

The partial sums correspond to repeated substitution in the equation

[ x=Tx+y. ]

Starting from the algebraic recurrence associated with this equation produces expressions of the form

[ x_N=\sum_{n=0}^{N}T^ny, ]

and the residual is

[ y-(I-T)x_N=T^{N+1}y. ]

Thus convergence of the operator series controls convergence simultaneously for every right-hand side, while convergence for an individual vector may occur under weaker conditions. This distinction links neumann series with fixed-point theory and with stationary methods for linear systems, but the operator identity remains independent of any particular computational interpretation.