External-Source Method in Mathematical Modeling

• External-Source Method is a framework that incorporates fixed, exogenous inputs into mathematical models to account for externally driven phenomena.
• It is applied across disciplines including epidemiology, quantum field theory, random matrix theory, and statistical integration.
• The method enhances model accuracy by quantifying external influences, enabling phase transition analysis and integrable reductions in complex systems.

The external-source method encompasses a class of theoretical, computational, and statistical techniques in which a physical, algebraic, statistical, or dynamical system is coupled explicitly to an externally specified input or source term. This formalism appears widely across disciplines—from mathematical epidemiology to quantum field theory, random matrix models, statistical estimation, and gauge/gravity theories. The recurring idea is the introduction, manipulation, and interpretation of an "external source"—a fixed function, field, current, or dataset—not determined by the system’s endogenous dynamics but entering as a known or parameterized non-autonomous input. The following sections outline the principal formulations, analytical structures, applications, and theoretical implications of the external-source methodology in contemporary research.

1. Analytical Structure and Mathematical Formulation

In model-building, the external-source method consists of modifying the system’s governing equations or probability measures via explicit source terms. The structure depends critically on context:

• Dynamical systems with external input: For epidemic models, the external-source term $E(t)$ represents the inflow of infected individuals from outside the modeled region. For instance, the time-delay epidemic system with external source (Chen et al., 2020) introduces $E(t)$ additively in

$\frac{dI}{dt} = \beta [I(t)-J(t)-G(t)] + E(t),$

modeling travel-driven seeding events.

• Statistical and machine learning settings: In data integration problems, the external-source method often enforces moment constraints or summary statistics sourced from external datasets—possibly biased—on a model fit to internal (individual-level) data. The generalized entropy balancing estimator (Morikawa et al., 13 Jun 2025) enforces

$\sum_{i=1}^n w_i\,g_j(X_i, Y_i) = m_j,$

where $m_j$ are external-source summary moments.

• Functional-integral and field-theoretic approaches: In quantum/gravitational field theory, the source $J(x)$ (or $T_{\mu\nu}(x)$) couples linearly as

$S[\phi, J] = S_0[\phi] + \int J(x)\,\phi(x)\,dx,$

or analogously for the gravitational field $h_{\mu\nu}$ and external stress-energy $T_{\mu\nu}$ (Manoukian et al., 2013). RG calculations and Schwinger-Dyson equations exploit functional derivatives with respect to these sources.

• Random matrix theory (RMT) and integrable models: The external source is typically realized as an additive shift to the Hamiltonian, e.g.,

$E(t)$0

leading to intricate modifications of the eigenvalue statistics and creating duality with higher-order or biorthogonal polynomial structures (Claeys et al., 2012, Desrosiers et al., 2013, Kimura, 2014, Wang et al., 23 Dec 2025).

2. Key Applications across Disciplines

2.1 Epidemiology and Dynamical Systems

The external-source term permits robust modeling of epidemic importation events under spatial heterogeneity. For the 2019-nCoV outbreak model, Chen et al. (Chen et al., 2020) demonstrated that only with $E(t)$1 corresponding to travel-driven importation does the time-delay model reproduce observed multi-peak outbreak curves and explains the amplification of the epidemic in regions with high mobility. Sensitivity analyses reveal substantial mitigation in final outbreak size with small increases in isolation rate $E(t)$2 even when subjected to exogenous inflows.

2.2 Quantum Field Theory and Statistical Mechanics

The inclusion of homogeneous or spatially varying sources is essential for defining generating functionals, computing linear and nonlinear response, and implementing nonperturbative approximations in field theory. For example, in the SO($E(t)$3) $E(t)$4 theory, the homogeneous external source enters in both the action and the self-consistency equations determining the critical line and renormalized masses (deLyra, 2014). In linearized gravity, the use of a non-conserved external source is critical for extracting the full structure—including Schwinger terms—of the graviton propagator (Manoukian et al., 2013).

2.3 Random Matrix Theory and Integrability

External-source RMT models generalize classical ensembles with significant consequences:

• Phase transitions and universality: The external-source leads to phenomena such as spiked eigenvalue transitions and the emergence of critical points where local eigenvalue statistics transition from Tracy-Widom/Airy to Gaussian or Pearcey/Boussinesq universality classes (Desrosiers et al., 2013, Wang et al., 23 Dec 2025).
• Multiple orthogonal polynomials and vector equilibrium: The presence of an external source complicates the equilibrium problem: for example, two-cut or constrained vector equilibrium determines the spectral density in quartic models with source (Bleher et al., 2010).
• Duality and integrable structure: Supermatrix and Hermitian models with external source admit duality relations and determinantal formulas directly involving the external source eigenvalues, which, upon tuning, generate intersection numbers of curves, Airy-type models, and connections to the Toda and KP hierarchies (Kimura, 2014, Brezin et al., 2018, Baik et al., 2012).

2.4 Statistical Estimation with Biased External Data

In statistical integration, the generalized entropy-balancing external-source estimator (Morikawa et al., 13 Jun 2025) utilizes summary moments from external (possibly biased) sources to impose convex moment-matching constraints on internal, individual-level data. Analytical and empirical results demonstrate double robustness: the estimator is consistent if either the response model or the external-source density ratio is correctly modeled, offering a protection mechanism not present in naive weighted or pooled estimators.

2.5 Computational and Algorithmic Settings

In stochastic iterative methods, external-source scenarios arise when the driving term of an operator equation is itself noisy (Monte Carlo estimated) (Makai et al., 2013). The algorithmic implications are nontrivial: correct variance estimation and convergence diagnostics require “block-frozen” (Type I) strategies where the random source realization is held fixed per run and only averaged in an outer loop. Methods that re-sample the source at each iteration (Type II) can fail to converge probabilistically.

3. Core Theoretical Properties and Threshold Phenomena

3.1. Thresholds and Phase Transitions

• Epidemic dynamics: The basic reproduction number under external sourcing, $E(t)$5, is unchanged by moderate external forcing, but external input can sustain persistent infection even when $E(t)$6 (Chen et al., 2020).
• Random matrices: External sources create phase transitions in eigenvalue statistics (e.g., subcritical/critical/supercritical regimes at soft edge) depending on the scale of the external eigenvalues relative to critical thresholds (Desrosiers et al., 2013, Wang et al., 23 Dec 2025). These transitions manifest in changes from Airy to Gaussian (or to higher-order Boussinesq-type) kernels in local spectral statistics.

3.2 Duality, Integrability, and Hierarchical Reduction

Determinantal and duality identities reduce high-rank external-source models to combinations of rank-1 cases. This is formalized by determinantal formulas built from single-spike kernels (e.g., Baik–Wang’s reduction via discrete KP hierarchy (Baik et al., 2012)), allowing full asymptotic solution of multi-spiked models by leveraging rank-1 analyses.

3.3 Gauge Ambiguity and Functional-Integral Implications

In NC field theory and quantum gravity, external-source extensions require working with non-conserved sources at the generating-functional level. The Schwinger term, nontrivial in gravity (Manoukian et al., 2013), or the ambiguity in the Seiberg-Witten map for NC QED (Adorno et al., 2011), are resolved only in the presence of arbitrary external source configurations, and only after full functional differentiation may physical constraints (like energy-momentum conservation) be imposed.

4. Algorithmic and Computational Aspects

Context	Source Implementation	Algorithmic Implication
Delay Epidemic Models	$E(t)$7 prescribed by travel data	Quantitative reproduction of multiple infective peaks; parameter fit via delay-differential solvers
Monte Carlo Iteration	$E(t)$8 sampled/stochastic	Stable and unbiased variance only with fixed-source per run. Re-sampling–per–iteration fails.
Data Integration	Summary moments $E(t)$9	Convex optimization for weights; double-robust estimation
Black Hole QNMs	Persistent/driven $\frac{dI}{dt} = \beta [I(t)-J(t)-G(t)] + E(t),$0	Only resonant/singular sources alter spectrum; regular sources shift amplitudes only

For example, in black-hole perturbation problems, an external periodic source (unlike an initial pulse) can introduce new quasinormal frequencies (dissipative singularities); otherwise the spectrum is preserved (Qian et al., 2020).

5. Limitations, Extensions, and Future Perspectives

5.1 Breakdown Regimes

• For operator matching in effective field theory, the external-source method is powerful at dimension-6—but not dimension-7 or higher, where operators with quark derivatives or multiple fermion bilinears cannot be matched simply by external classical sources; spurion analysis is required instead (Li et al., 3 Jul 2025).
• In external-source MC problems, convergence-in-probability can be prohibitively low for large system size unless statistical noise is aggressively reduced or convergence criteria relaxed (Makai et al., 2013).

5.2 Generalizations

• In random matrix theory, the external-source paradigm encompasses a wide variety of ensembles, including those with equispaced source spectra (biorthogonal ensembles), supermatrix extensions, and models yielding higher-order/gravitational critical universality (Wang et al., 23 Dec 2025, Kimura, 2014, Brezin et al., 2018, Claeys et al., 2012).
• In dialogue generation with LLMs, factuality-diversity trade-offs are controlled in decoding by dynamically exposing or masking external knowledge sources at each token, based on confidence metrics (Yang et al., 2024).

5.3 Further Theoretical and Applied Directions

Current developments leverage the external-source method for phase transition analysis in high-dimensional statistics, universality proofs in RMT, precision modeling in epidemiology and physics, and robust integration of heterogeneous datasets. A plausible implication is that increasingly complex coupled stochastic and data-fusion problems will adopt structured external-source strategies for both modeling accuracy and analytical tractability.

6. Summary Table: Key Forms and Functions of the External-Source Method

Domain	Source Term / Mechanism	Core Result / Role	Reference
Epidemics	$\frac{dI}{dt} = \beta [I(t)-J(t)-G(t)] + E(t),$1: travel influx	Accurately models multi-modal outbreaks, quantifies isolation gains	(Chen et al., 2020)
Field theory	$\frac{dI}{dt} = \beta [I(t)-J(t)-G(t)] + E(t),$2	Generates functional derivatives, critical lines, renormalized masses	(1402.18631311.2390)
Random matrices	$\frac{dI}{dt} = \beta [I(t)-J(t)-G(t)] + E(t),$3: spectrum shift	Transitions in spectral kernel universality, integrable structures	(1212.37681306.4058Wang et al., 23 Dec 2025)
Data integration	External moments $\frac{dI}{dt} = \beta [I(t)-J(t)-G(t)] + E(t),$4	Improves estimator efficiency, double robustness	(Morikawa et al., 13 Jun 2025)
MC simulation	Stochastic $\frac{dI}{dt} = \beta [I(t)-J(t)-G(t)] + E(t),$5	Precise variance, correct convergence only with “block-freeze”	(Makai et al., 2013)
Black-hole QNMs	Driven $\frac{dI}{dt} = \beta [I(t)-J(t)-G(t)] + E(t),$6	Regular sources preserve, resonant create, new QNMs	(Qian et al., 2020)
Dialogue generation	Factual mask at each token	Dynamic trade-off: factuality vs. diversity in response generation	(Yang et al., 2024)

7. Concluding Perspective

The external-source method is a unifying analytical tool across theoretical and applied science for encoding external influences, constraints, or datasets. Its rigor, tractability, and ability to generate integrable reductions or robust statistical estimators explain its ubiquity and continued development in modern mathematical modeling and statistical inference. Recent work demonstrates its central importance for understanding phase transitions, universality, and the behavior of complex, externally modulated systems.