Neumann Series: Convergence, Computation & Applications

• Neumann Series is an infinite expansion representing the inverse of operators (I-A) under specific spectral restrictions, widely used in matrix analysis and integral equations.
• It accelerates computations by enabling truncated approximations with predictable error bounds and efficient iterative methods in various numerical and quantum applications.
• Its practical applications extend to regression adjustments, graph signal processing, and special function representations, providing versatile analytic and computational benefits.

The Neumann series is a central analytic representation in both operator theory and applied mathematics, providing a formal infinite expansion for the inverse of linear operators of the form $I-A$ under appropriate spectral conditions. Classically, the Neumann series and its truncated or specialized variants appear in matrix analysis, integral equations, numerical algorithms (notably in iterative linear solvers), quantum error mitigation, regression adjustment, and series expansions for special functions. These applications leverage its convergence properties, hereditary structure for power sums, and, in advanced cases, its universality under random matrix fluctuations.

1. Fundamentals and Convergence Criteria

A Neumann series is defined for a bounded linear operator $A$ (on a Banach or Hilbert space, or, by extension, a matrix) as the expansion: $(I - A)^{-1} = \sum_{k=0}^\infty A^k,$ which converges in operator norm if $\|A\| < 1$ or, more generally, if the spectral radius $\rho(A) < 1$ (Lemma 2.1, (Zhang et al., 2019)). For a linear system $(I-A)x = b$, the exact solution can be represented as $x = (I-A)^{-1}b = \sum_{k=0}^\infty A^k b$, and the iterative process $x_{k} = Ax_{k-1} + b$ converges to $x$ as $k \to \infty$ under the same spectral restriction [(Theorem 2.2, (Zhang et al., 2019))].

For integral operators of Volterra type in Banach spaces, absolute convergence is guaranteed even when $A$0 is not strictly less than 1, due to the Volterra structure enforcing a zero spectral radius. The Neumann–Picard iteration for the Volterra equation $A$1 converges in the relevant Bochner-Lebesgue space for all $A$2, providing unique solutions to integral and differential equations [(Theorems 4-6, (Mera, 2011))].

Table 1: Core Convergence Criteria

Context	Spectral Criterion	Main Reference
Matrix/operator inverse	$A$3 or $A$4	(Zhang et al., 2019)
Volterra integral operator	Always convergent, spectral radius 0	(Mera, 2011)
Quantum channel inversion	$A$5 in operator norm	(Wang et al., 2021Wang et al., 2021)

2. Accelerated Computation and Truncation

Computing high-degree truncations of the Neumann series is crucial for practical inversion. Factoring the geometric series polynomial enables substantial reduction in computational complexity, especially for large $A$6 [(Dimitrov et al., 2017)]. Binary, ternary, and higher-order (e.g., basis size 5) decompositions reduce the cost of evaluating $A$7 from $A$8 multiplications (binary) to about $A$9 for optimal mixed-base algorithms. Fast recursive factorizations and residue-based strategies allow for practical matrix inversion and are critical in real-time applications (e.g., wireless communications, graphics rendering).

Truncation analysis is central in numerical applications and quantum error mitigation. Truncating after $(I - A)^{-1} = \sum_{k=0}^\infty A^k,$0 terms leads to an error bound $(I - A)^{-1} = \sum_{k=0}^\infty A^k,$1, and practical error mitigation leverages exponential decay in bias with increasing truncation order, the overhead depending only on the noise resilience parameter and not on system size [(Wang et al., 2021, Wang et al., 2021, Dimitrov et al., 2017)].

3. Neumann Series in Applied Numerical Linear Algebra

The Neumann series serves as an analytic foundation for iterative linear equation solvers and preconditioning in numerical algorithms [(Thomas et al., 2021, Zhang et al., 2019)]. In the context of GMRES, replacing explicit triangular solves by truncated Neumann expansions ($(I - A)^{-1} = \sum_{k=0}^\infty A^k,$2 is approximated by $(I - A)^{-1} = \sum_{k=0}^\infty A^k,$3) achieves backward stability and maintains convergence rates within quantified error bounds (Paige et al., 2006). Polynomial smoothers for algebraic multigrid preconditioners and iterative ILU schemes use the Neumann expansion to convert otherwise sequential triangular solves into efficiently parallelizable sparse matrix-vector multiplies. Dense random matrix regimes have been analyzed probabilistically: the number of required Neumann iterations to reach a residual threshold is a function solely of the scaled extreme eigenvalues, leading to explicit distributions for iteration counts under various random matrix ensembles [(Zhang et al., 2019)].

4. Quantum Error Mitigation via Truncated Neumann Series

Quantum computing implementations are hindered by Markovian gate and measurement noise, typically modeled by stochastic matrices or Pauli transfer matrices. Both gate error mitigation (GEM) and measurement error mitigation (MEM) utilize truncated Neumann expansions to approximate the inverse of the noise channel. For a noise map $(I - A)^{-1} = \sum_{k=0}^\infty A^k,$4 with $(I - A)^{-1} = \sum_{k=0}^\infty A^k,$5, the ideal expectation value is expressed as a linear combination of noisy measurements at multiple depths, using binomial coefficients derived from the series expansion. The key advantages are device- and system-size-independence of the mitigation overhead, scalability, and exponentially decaying systematic bias as a function of truncation order ($(I - A)^{-1} = \sum_{k=0}^\infty A^k,$6) [(Wang et al., 2021, Wang et al., 2021)].

5. Neumann Series Expansions for Differential Equations and Special Functions

Neumann Series of Bessel Functions

General Neumann series of Bessel functions (also: Schlömilch–Neumann expansions) take the form

$(I - A)^{-1} = \sum_{k=0}^\infty A^k,$7

with broad applicability: integral representations, asymptotic analysis, and uniform convergence across families of special functions including Kelvin, Lommel, and confluent hypergeometric functions [(Micheli, 2017, Sanford, 2016)]. For Sturm–Liouville and perturbed Bessel equations, linearly independent solutions can be constructed as Neumann series involving either cylindrical or spherical Bessel functions with coefficients calculated from recursive relations involving transmutation kernels, SPPS formal powers, or Fourier–Legendre expansions [(Kravchenko et al., 2016, Kravchenko et al., 2016, Kravchenko et al., 2015)].

These representations have the property that the error in a truncated series expansion can be made independent of the spectral parameter, thus facilitating eigenvalue computation and efficient numerical algorithms for boundary value and spectral problems.

Neumann-Type Series for Modified Bessel Functions

In harmonic analysis, generalized Neumann series involving modified Bessel functions arise, notably in the analysis of radial Dunkl operators for dihedral groups. Explicit multi-parameter expansions exist, indexed by representation parameters and symmetric polynomials (Gegenbauer), and reductions to Horn hypergeometric functions reveal intricate algebraic structures [(Deleaval et al., 2017)].

Table 2: Applications of NSBF Expansions

Equation/Class	Neumann Series Structure	Reference
Sturm–Liouville / Schrodinger	Sums over spherical/cylindrical Bessel	(Kravchenko et al., 2016, Kravchenko et al., 2015)
Perturbed Bessel Equation	Fourier–Legendre → spherical Bessels	(Kravchenko et al., 2016)
Sine and Cosine Integrals	Integer-order Bessel series	(Sanford, 2016)

6. Extensions: Regression Adjustment and Graph Signal Processing

Recent work extends the Neumann series framework to statistical estimation under randomization, specifically for regression adjustment in high-dimensional randomized experiments [(Song, 11 Nov 2025)]. Decomposing the design-induced covariance matrix’s inverse as a Neumann series reveals a hierarchy of correction terms with explicit tail control, enabling new normality results under relaxed dimensionality regimes (e.g., admissible dimension rates for OLS adjustment are increased incrementally by including higher-order Neumann corrections).

In graph signal processing, the Neumann series enables efficient A-optimal sampling and robust signal reconstruction, especially for the inversion of principal submatrices within ideal filter spaces. Chebyshev polynomial approximations further accelerate computation by avoiding explicit eigen-decomposition [(Wang et al., 2018)].

7. Integral Representations and Generalizations

The Neumann series, especially in function theory, admits a single-integral master representation via a universal kernel and the generating function of coefficients, as shown for Bessel series by De Micheli [(Micheli, 2017)]. This unifies explicit expansion, analytic continuation, and asymptotic analysis across a wide class of special functions.

Integral representations and closure properties of Neumann series are also critically used in deriving explicit formulas for higher-order polynomials (e.g., Gegenbauer), orthogonal polynomials, and in probabilistic analysis of spectral algorithms [(Deleaval et al., 2017)].

---

References:

• (Zhang et al., 2019): A Probabilistic Analysis of the Neumann Series Iteration
• (Mera, 2011): Convergence of the Neumann series for the Schrödinger equation and general Volterra equations in Banach spaces
• (Wang et al., 2021): Mitigating Quantum Errors via Truncated Neumann Series
• (Wang et al., 2021): Measurement Error Mitigation via Truncated Neumann Series
• (Thomas et al., 2021): Neumann Series in GMRES and Algebraic Multigrid Smoothers
• (Song, 11 Nov 2025): Neumann-series corrections for regression adjustment in randomized experiments
• (Dimitrov et al., 2017): On the Computation of Neumann Series
• (Wang et al., 2018): A-Optimal Sampling and Robust Reconstruction for Graph Signals via Truncated Neumann Series
• (Kravchenko et al., 2016): A Neumann series of Bessel functions representation for solutions of Sturm-Liouville equations
• (Kravchenko et al., 2016): A Neumann series of Bessel functions representation for solutions of perturbed Bessel equations
• (Kravchenko et al., 2015): Representation of solutions to the one-dimensional Schrödinger equation in terms of Neumann series of Bessel functions
• (Micheli, 2017): Integral representation for Bessel's functions of the first kind and Neumann series
• (Sanford, 2016): Two Neumann Series Expansions for the Sine and Cosine Integrals
• (Deleaval et al., 2017): On a Neumann-type series for modified Bessel functions of the first kind