Новый метод описания андерсоновской локализации через модальную декомпозицию

1367-2630/28/9/093502 Abstract Anderson localization in one-dimensional disordered media is traditionally characterized through transfer-matrix growth and Lyapunov exponents in the infinite-chain limit. Although mathematically rigorous, this approach obscures the underlying physical mechanism and requires extrapolation for finite systems. We develop an alternative formulation rooted in the modal decomposition of finite disordered chains and random eigenvalues. Starting from time-domain impulse response analysis, we show that disorder-induced eigenvalue scatter causes each modal contribution to acquire a random phase deviation that grows linearly in time, producing a Gaussian decay of the ensemble-averaged modal amplitude with a characteristic decoherence time inversely proportional to both the disorder strength and the eigenvalue variance. This modal phase decoherence manifests in the frequency domain as exponential decay of transmission amplitude, which we express exactly for any finite chain length via a spectral product over eigenvalue detunings derived from Cramer’s rule. For Gaussian disorder, the ensemble-averaged localization factor is obtained in closed form using Dawson’s integral; for uniform disorder an exact elementary expression is derived. We prove rigorously that the localization factor is independent of the disorder distribution to leading order, depending only on the variance, establishing Anderson localization as a second-moment statistical process. The continuum limit recovers the classical transfer-matrix prediction for the coupling-parameter dependence, and Monte Carlo simulations confirm that the disorder-to-coupling ratio is the fundamental dimensionless parameter governing weak-to-strong localization transitions, with the scaled localization factor collapsing onto a universal master curve. The formulation provides exact finite-system results without infinite-chain extrapolation and offers direct physical insight into the decoherence mechanism underlying Anderson localization. Export citation and abstract BibTeX RIS Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 license. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. 1. Introduction Anderson localization, the absence of wave diffusion in disordered media, is one of the most far-reaching consequences of interference in random systems [ 1 ]. Originally predicted for electrons in disordered lattices, the phenomenon is universal: it governs the transport of classical waves whenever scattering from disorder is strong enough to destroy phase coherence over the system length. Experimental confirmations span electromagnetic waves in random dielectrics [ 2, 3 ], matter waves in optical speckle potentials [ 4 ], and ultrasound in three-dimensional elastic networks [ 5 ]. Anderson localization of elastic and flexural waves in finite disordered structures has been confirmed experimentally [ 6 ] and the transition in three-dimensional elastic solids established numerically [ 7 ], demonstrating that the phenomenon remains an active area of classical wave physics. The scaling theory of localization [ 8 ] established that all states are localized in one and two dimensions for arbitrarily weak disorder, making one-dimensional systems the canonical setting for both analytical treatment and experimental investigation. The standard analytical framework for one-dimensional localization rests on transfer matrices and Lyapunov exponents [ 9 – 12 ]. The chain is decomposed into a product of matrices relating amplitudes at adjacent sites; the localization length emerges as the inverse Lyapunov exponent of this product in the infinite-chain limit [ 13 ]. For phononic and structural systems, Pierre [ 14 ] applied this framework to derive closed-form expressions for the localization factor as a function of excitation frequency, disorder strength, and inter-site coupling in both weak and strong disorder regimes, establishing the essential role of the disorder-to-coupling ratio in governing the transition between the two regimes. These results have been confirmed extensively in numerical studies of disordered chains [ 15 – 18 ]. A thorough account of anomalous localization in one-dimensional systems, including the role of correlated disorder and finite-size corrections, is given in [ 19 ]. The transfer-matrix approach is rigorous and delivers the correct asymptotic scaling, but it has two limitations that motivate the present work. The first limitation is mechanistic. The Lyapunov exponent is a property of an infinite product of random matrices; it characterizes the rate of localization and leaves the mechanism, in terms of the vibrational modes of the chain, unexplained. The question of why disorder converts propagating Bloch waves into spatially localized modes, phrased in the language of modal superposition and phase relationships in place of matrix products, lies outside the reach of the transfer-matrix formalism. The second limitation is practical. Every physical phononic system is finite, and the Lyapunov exponent is an infinite-system limit that must be extrapolated from numerical data. An exact analytical expression for the disorder-averaged transmission amplitude of a finite disordered chain, requiring only the eigenvalue spectrum and free of infinite-system construction, remains outstanding. The growing importance of disordered phononic metamaterials, where manufacturing tolerances are unavoidable and system sizes are finite [ 20 ], makes both limitations practically consequential. Characterizing how disorder affects wave transmission requires analytical tools that work directly with finite systems and connect to the modal physics of the structure. Random eigenvalue problems provide precisely such a framework [ 21 – 24 ]: eigenvalue scatter induced by parameter randomness is the natural object of study, and its statistical properties can be characterized in closed form for a broad class of disorder distributions. This paper shows that Anderson localization in one-dimensional disordered chains is the spatial manifestation of modal phase decoherence. Working within the modal decomposition of finite chains, we establish that disorder-induced eigenvalue scatter causes the modal contributions to the impulse response to lose phase coherence across the disorder ensemble: each mode acquires a random phase deviation, whose variance grows quadratically with time, producing a Gaussian decay of the ensemble-averaged modal amplitude with a characteristic decoherence time. In the frequency domain, the same eigenvalue scatter randomizes the resonance denominators in the modal expansion of the harmonic response, and the accumulated destructive interference over all modes causes the transmission amplitude to decay exponentially with distance. This connection between temporal phase randomization and spatial amplitude decay, both driven by the identical eigenvalue perturbation, is the central result of the paper. The mechanism is an exact classical analogue of quantum decoherence: coupling to a random environment destroys the phase relationships that sustain coherent transport. Our formulation yields six results. Starting from an exact expression for the transmission amplitude via Cramer’s rule [ 25 ], we derive the localization factor as a spectral sum over eigenvalue detunings, valid for any finite chain length without infinite-system extrapolation. For Gaussian disorder, the ensemble average of this spectral sum is evaluated exactly in closed form using Dawson’s integral [ 26 ]; for uniform disorder an exact elementary expression is derived. We prove rigorously that the disorder-averaged localization factor is independent of the disorder distribution to leading order, depending only on the variance, a signature of second-moment statistical processes, with distribution-dependent corrections entering only at fourth order in the disorder strength. The disorder-to-coupling ratio is identified analytically as the fundamental dimensionless parameter governing weak-to-strong localization transitions, and the scaled localization factor is shown to collapse onto a universal master curve. In the continuum limit, a Riccati map analysis of the governing difference equation yields an explicit closed-form localization factor whose parametric dependence on disorder variance and coupling confirms consistency between the finite-chain spectral formulation and the classical infinite-chain limit, without appeal to external results. Finally, the decoherence mechanism itself is shown to be distribution-independent: the phase coherence factor depends only on the variance of the disorder-induced phase deviation, regardless of the distribution shape, reinforcing the second-moment character of the phenomenon. The paper is organized as follows. Section 2 introduces the monoatomic phononic chain, derives the exact eigensolution of the ordered system, and specifies the disorder model. Section 3 establishes the modal phase decoherence mechanism from time-domain impulse response analysis and connects it to harmonic transmission decay. Section 4 derives the exact spectral representation of the localization factor via Cramer’s rule and establishes the ordered-chain baseline. Section 5 performs the disorder averaging for Gaussian disorder using Dawson’s integral, derives the weak-disorder asymptotic expansion, and obtains the continuum-limit localization factor via a Riccati map analysis of the governing difference equation. Section 6 proves distribution independence, derives the closed-form result for uniform disorder, and identifies as the universal governing parameter. Section 7 discusses the implications of the decoherence framework and directions for extension to multi-branch and higher-dimensional systems. 2. System model and modal analysis 2.1. Monoatomic chain: dispersion relationship The discrete periodic medium under consideration is modelled as a one-dimensional monoatomic phononic crystal, as illustrated in figure 1. Each unit cell consists of a concentrated mass coupled to its adjacent neighbors via a horizontal spring with stiffness. To account for local resonance or substrate effects, each mass is further connected to a rigid ground through a grounding spring, where the index identifies the th lattice site. This configuration allows for the study of wave propagation and the formation of band gaps within the lattice. The equation of motion for the th oscillator reads subject to fixed boundary conditions. We nondimensionalize using the natural frequency of an isolated grounded oscillator,, and introduce dimensionless quantities Here is the dimensionless excitation frequency and is the coupling parameter quantifying the relative strength of inter-oscillator coupling to individual oscillator stiffness. For harmonic excitation at frequency, substituting into the dimensionless equation of motion yields For an infinite ordered chain, we seek Bloch wave solutions where is the wavenumber, is the lattice constant, and is the wave amplitude. Substituting equation ( 4 ) into equation ( 3 ) gives which upon rearrangement, yields the dispersion relation The passband occupies, bounded below by the isolated oscillator frequency at the zone center (, all masses moving in phase) and above by the zone-edge frequency (, adjacent masses oscillating out of phase). Stopbands exist for (below the grounding frequency, where restoring forces from grounded springs dominate and waves decay evanescently) and (above the zone edge, where the system ceases to support propagating waves). The coupling parameter controls passband width and wave transmission characteristics: weak coupling ( ) produces a narrow passband concentrated near, rendering the system nearly a collection of isolated oscillators with minimal energy transfer, and strong coupling ( ) yields a wide passband extending to, promoting efficient wave propagation across the chain. This coupling dependence has direct implications for disorder-induced localization, as will be demonstrated in subsequent sections. For a finite chain of masses with fixed boundaries, the continuous spectrum ( 6 ) is discretized, admitting only wavenumbers for. Substituting into equation ( 6 ) gives the eigenfrequencies The corresponding mode shapes are, forming an orthonormal basis. These deterministic eigensolutions provide the reference against which disorder-induced localization will be characterized. Figure 2 shows the dispersion curves of the monoatomic chain for different values of the coupling parameter. As increases, the upper bound of the passband shifts to higher frequencies and the lower bound remains fixed at, corresponding to the natural frequency of an isolated oscillator. Consequently, the passband width increases approximately linearly with, reflecting stronger inter-oscillator coupling and more efficient wave propagation across the chain. In the weak coupling limit ( ), the band becomes narrow, and the system behaves as nearly decoupled oscillators, whereas for large the dispersion curve becomes significantly steeper, indicating enhanced wave mobility. This dependence on plays a central role in determining the susceptibility of the system to disorder-induced localization, as will be discussed in subsequent sections. 2.2. Ordered chain: deterministic modal structure For the ordered chain ( ), the eigenvalue problem admits an exact analytical solution. The governing difference equation obtained from equation ( 3 ) is subject to fixed boundary conditions. We seek separable solutions of the form which automatically satisfy the boundary condition at. Substituting ( 9 ) into ( 8 ) gives which is the discrete dispersion relation. The second boundary condition enforces Substituting ( 11 ) into ( 10 ) yields the exact eigenvalues which correspond to discrete sampling of the continuous dispersion relation ( 6 ). The corresponding normalized eigenvectors are obtained by enforcing unit norm, giving which satisfy the orthonormality condition The eigenfrequency spectrum ( 12 ) lies entirely within the passband with the lower bound corresponding to the in-phase motion of all masses and the upper bound to the out-of-phase (zone-edge) oscillation. Physically, the mode shapes ( 13 ) represent standing waves in the finite chain: the th mode contains half-wavelengths across the lattice and internal nodes. The coupling parameter controls the spread of the eigenvalues, and the boundary conditions alone determine the spatial structure of the modes. This deterministic eigensolution forms the zeroth-order reference for subsequent perturbation analysis of disordered chains, where disorder induces eigenvalue shifts and modal phase perturbations relative to equations ( 12 ) and ( 13 ). 2.3. Disorder model We introduce structural disorder through random variations in the grounding spring stiffnesses. The stiffness of the grounding spring at site is modeled as where is the nominal stiffness, is the dimensionless disorder strength, and are independent random variables characterizing the spatial fluctuations. The random variables are assumed to satisfy where denotes the ensemble average over disorder realizations. The zero-mean condition ensures that the average grounding stiffness remains, and the unit variance convention normalizes the disorder strength to the single parameter. Statistical independence reflects the assumption that disorder at different sites is uncorrelated, as would arise from independent manufacturing tolerances or local material variations. Two canonical distributions will be considered: Gaussian disorder:, appropriate for disorder arising from accumulated small random perturbations via the central limit theorem. Uniform disorder:, representing bounded manufacturing tolerances with equal probability across a specified range. The bounds ensure unit variance. In dimensionless form, the equations of harmonic motion can be written as the matrix equation where is the displacement vector, is the nondimensional force vector with excitation at the first mass, is the identity matrix, and is the system matrix. For the monoatomic chain with grounded oscillators, the matrix has the tridiagonal structure The diagonal elements represent the combined restoring forces from the grounded spring (unity) and the two coupling springs ( ), plus the random perturbation ( ). The off-diagonal elements represent the coupling between adjacent oscillators. This can be decomposed as where is the ordered tridiagonal matrix (obtained by setting in ( 19 )) and is the diagonal disorder matrix. The diagonal structure of reflects the local nature of grounding spring disorder: fluctuations at site affect only the th degree of freedom directly, with coupling to neighboring sites mediated indirectly through the off-diagonal terms in. This perturbation structure permits analytical treatment via standard eigenvalue perturbation theory, as will be developed in subsequent sections. 3. Modal phase decoherence in the time domain This section establishes the fundamental mechanism underlying Anderson localization: disorder-induced eigenvalue scatter causes modal phases to decohere, preventing coherent wave transmission. We develop this concept from first principles using time-domain modal decomposition. 3.1. Modal decomposition of impulse response Consider an impulsive excitation applied at site of the chain, where is the Dirac delta function and is the Kronecker delta selecting the excitation location. This impulse imparts unit momentum to mass at, establishing initial conditions and (with all other velocities zero). The subsequent free vibration can be decomposed into the modal basis defined by equation ( 13 ). Expanding the displacement vector as, where denotes the eigenvector with components and are modal coordinates, the equations of motion decouple into independent modal oscillators, The initial conditions in modal coordinates, obtained by projecting the physical initial conditions onto the modal basis using orthonormality ( 14 ), are Solving equation ( 23 ) with these initial conditions gives The response at site is obtained by transforming back to physical coordinates, This exact result shows that the impulse response is a coherent superposition of sinusoidal contributions, each oscillating at its natural frequency with amplitude weighted by the product. The modal participation factor quantifies how effectively mode couples the excitation at site to the response at site: modes with significant amplitude at both locations contribute strongly, and modes with nodes near either location contribute weakly. For the ordered chain with eigenfrequencies given by ( 12 ), all modal contributions in ( 26 ) begin in phase at. Each mode has zero displacement and phase at the initial instant. As time evolves, the different frequencies cause the modal phases to advance at different rates, but this phase progression is deterministic and identical across all disorder realizations for the ordered system. The coherent superposition persists for all time, producing the characteristic beating patterns associated with multi-frequency oscillations in linear systems. 3.2. Disorder-induced eigenvalue scatter Disorder in the grounding spring stiffnesses perturbs the eigenvalues from their deterministic values given in ( 12 ). Treating the disorder matrix as a small perturbation with strength, standard first-order perturbation theory (derived in appendix A ) yields where is the deterministic eigenvalue from ( 12 ) and is the random perturbation variable The random variable is a weighted sum of the independent disorder variables, with weights given by the squared mode shape amplitudes. Since and for, it follows immediately that where the variance is Evaluating the sum using trigonometric identities (see appendix A ), the asymptotic result for large is The asymptotic step is in fact avoidable. The two cosine sums generated by the quartic sine identity take the exact value, which is sharper than the bound used above, so the sum in ( 30 ) has a closed elementary value. As shown in appendix A, with for the exceptional mode, which exists only for odd. The variance is exactly independent of mode index, so band-edge modes carry the same scatter as the rest and share the same treatment. Equation ( 31 ) overestimates the variance by, which is at, and the standard deviation by: exactly against. The exact value is used throughout; ( 31 ) is retained for order-of-magnitude estimates and for the scaling argument of section 6.4. This result has important physical implications. First, the variance is independent of the mode index to leading order, meaning all modes experience comparable eigenvalue scatter. Second, the variance decreases as: larger systems exhibit smaller relative eigenvalue fluctuations because the modal weights distribute the disorder over more degrees of freedom. Third, combining ( 27 ) and ( 29 ), each eigenfrequency has variance, creating frequency uncertainty that scales linearly with the disorder strength. The validity condition for ( 27 ) is. Its spectral dependence enters through, which by ( 12 ) spans, so the condition binds hardest at the bottom of the passband. With ( 32 ), which at is for and for. A three-standard-deviation excursion shifts an eigenvalue by and. Both disorder levels used here lie inside the perturbative regime. The controlling parameter is therefore, in place of alone or of the disorder-to-coupling ratio of section 6.4. The two measure different things: is how far the spectrum moves, is how strongly the resulting scatter localizes. A chain at of order one half can be strongly localized with accurately described eigenvalues, which is why the collapse of figure 9 holds over its full tested range,. The bound ( 33 ) falls as, so accuracy improves with system size; for a few dozen oscillators the limit is, where the excursion reaches. Equation ( 33 ) measures the shift against the eigenvalue. A second, independent criterion measures it against the mean level spacing, because nondegenerate perturbation theory also neglects eigenvector mixing, and that neglect is controlled by the spacing. The ratio is of ( 83 ). The two differ: at, the first gives and is satisfied, and at exceeds unity. Figure 3 depends only on the first; the closed-form localization factor depends on both, as examined in subsection 5.1. In figure 3, the relative frequency scatter, where denotes the standard deviation of the natural frequency, is compared across three predictions: Monte Carlo simulations, first-order perturbation theory (equation ( 30 )), and the asymptotic approximation (equation ( 31 )). The results in figure 3 demonstrate excellent agreement between all three predictions across the spectrum. The first-order perturbation formula (equation ( 30 )) closely tracks the Monte Carlo results for both disorder levels, confirming the accuracy of the analytical variance expression. The asymptotic approximation (equation ( 31 )), which predicts mode-independent variance, also captures the average behavior accurately. The strong agreement validates the perturbative framework and confirms that frequency scatter is well described by a second-moment statistical process, providing direct support for the phase decoherence mechanism underlying Anderson localization. The are correlated across modes. Every weight vector in ( 28 ) sums to unity, so all modes share a component proportional to the sample mean of the disorder. The cosine identities that give ( 32 ) also give for, derived in appendix A, hence a correlation coefficient of exactly for pairs whose indices both differ from the exceptional mode, and for pairs that include it. For the weight vectors coincide, so identically. Every result here is unaffected, for a structural reason. The average ( 76 ) is a sum of modal expectations, and linearity of expectation uses only the marginal of each. Correlation enters the variance of, through, where at it would be a leading-order effect. That variance lies outside the present scope and is identified in section 7 as the next step. 3.3. Phase evolution and modal decoherence We now establish the fundamental mechanism by which disorder prevents wave transmission: eigenvalue scatter causes modal contributions to lose phase coherence across the disorder ensemble. For a single realization, the impulse response at site is given by ( 26 ) with perturbed eigenfrequencies from ( 27 ). To first order in, the perturbed frequency is where we have used the binomial approximation valid for. The phase of mode at time is therefore where the first term is the deterministic phase advance and the second is the disorder-induced phase deviation This is a zero-mean random variable with variance Different disorder realizations produce different phase deviations, causing modal contributions to scatter in phase across the ensemble. The single-mode contribution to ( 26 ) is where has been expanded to first order. Using and taking the ensemble average, the linear terms in vanish since, giving to leading order in, where the phase coherence factor quantifies the degree to which modal phases remain aligned across the ensemble. For Gaussian disorder,, so is Gaussian with variance ( 37 ). Using the standard identity for a zero-mean Gaussian, Defining the modal decoherence time this condenses to and the ensemble-averaged impulse response is Each modal contribution decays with a Gaussian envelope governed by: at all realizations are in phase ( ); as grows, phase scatter accumulates linearly via ( 36 ); at destructive interference suppresses the ensemble average; for the modal contribution vanishes completely. The decoherence time is inversely proportional to disorder strength and eigenvalue variance, and directly proportional to modal frequency: stronger disorder accelerates decoherence, and higher-frequency modes resist it because their faster phase advance requires larger deviations to produce destructive interference. Figure 4 validates equation ( 45 ) against Monte Carlo simulations. The time axis is normalized by the fundamental period and the response by the static displacement of the first mass, providing an amplitude scale independent of dynamic amplification. The left subfigures confirm that the single-mode ensemble average follows the Gaussian decay in ( 45 ) with high precision, directly validating the phase decoherence mechanism. The right subfigures show that first-order theory captures the qualitative decay of the full response but underestimates the persistence of oscillations; this is expected because the model retains unperturbed eigenvectors, whereas disorder also induces weak inter-modal coupling that redistributes energy at longer times. The single-mode agreement is the essential result: it confirms that eigenvalue-driven phase scatter is the operative mechanism, and the full-response discrepancy is a quantitative limitation of first-order perturbation theory that leaves the spectral localization results of section 4 onwards intact. The same eigenvalue scatter that randomizes modal phases in the time domain also randomizes modal amplitudes in the frequency domain. Under harmonic excitation at frequency, the modal amplitude of mode is proportional to: disorder perturbs by, so the denominator fluctuates across realizations. For modes near resonance, where, these fluctuations are of the same order as the detuning itself, producing large amplitude scatter and destructive interference in the ensemble average. The accumulated destructive interference over the modes of the chain is what causes wave transmission to decay with distance, the spatial manifestation of modal phase decoherence, derived precisely in the following subsection. 3.4. Spatial manifestation: transmission decay The temporal phase decoherence established in the previous subsection has a direct spatial manifestation: the same eigenvalue scatter that randomizes modal phases in time also randomizes modal amplitudes in the frequency domain, causing the ensemble-averaged transmission to decay exponentially with distance from the excitation point. Consider harmonic excitation at dimensionless frequency applied at site. The steady-state displacement at site satisfies where is the force vector with unit amplitude at site. Expanding in the modal basis and projecting onto mode using orthonormality ( 14 ), so the modal amplitude is and the harmonic response at site is Each mode contributes with weight inversely proportional to its detuning from the excitation frequency: modes near resonance dominate, and far-detuned modes contribute weakly. Defining the deterministic detuning and substituting the first-order perturbation from ( 27 ), the modal denominator becomes, a random variable that fluctuates across disorder realizations. The mechanism by which this amplitude fluctuation produces spatial decay is the frequency-domain counterpart of temporal phase decoherence. For a near-resonant mode with, the disorder perturbation is comparable to the detuning itself, so different realizations place the eigenvalue on opposite sides of, producing modal amplitudes of alternating sign and large magnitude. When summed over the ensemble, these contributions interfere destructively, suppressing. The suppression grows with the propagation distance because transmitting energy to a distant site requires phase coherence to be maintained across an increasing number of modal contributions simultaneously; disorder-induced eigenvalue scatter erodes this coherence progressively, converting what would be propagating Bloch waves into exponentially decaying localized modes. Figure 5 demonstrates this mechanism directly by showing the typical transmission envelope as a function of site index for varying disorder strengths and coupling parameters. The quantity plotted is the typical, or geometric, transmission amplitude as distinct from the arithmetic mean. The distinction matters in one dimension: the arithmetic average of the transmission is dominated by rare realizations in which a resonant mode aligns with the excitation, and decays at a rate different from the typical one. Every analytical result here is a typical quantity by construction. Definition ( 60 ) takes the logarithm before averaging, so ( 76 ) is, in place of, giving. By ( 62 ), is a sum of logarithmic terms, so concentrates about its mean as grows, whereas retains its spread. For weak disorder ( ), the transmission envelope is essentially flat, confirming that the ordered-chain Bloch waves survive weak perturbation. For moderate disorder ( ), the envelope decays exponentially with site index, with a rate that increases as the coupling parameter decreases, consistent with the observation from ( 37 ) that smaller narrows the eigenvalue band, reducing the detunings and thereby amplifying the sensitivity of near-resonant modes to disorder. The even-site and odd-site decay rates differ through a fitting artifact in place of a parity effect, and the printed values establish this. At with,, so in all four subfigures. The residual carrier modulating the envelope is sampled at different phases on the two sublattices, biasing the least-squares slopes in opposite directions by a fixed amount and setting a floor on the resolvable decay rate. In (a), (b) and (d), where ( 99 ) predicts, and, the fitted pairs straddle that floor and differ by, and. In (c), where the prediction is an order of magnitude larger, they agree to, or of their own value. A parity effect would persist as the decay rate rises; a floor-limited artifact switches off, which is what the figure shows. Subfigure (c) benchmarks the formulation against classical transfer-matrix theory. With, and, so, the established weak-disorder Lyapunov exponent for the one-dimensional Anderson model [ 10, 11, 19 ], recovered independently as ( 99 ) and applied to phononic chains by Pierre [ 14 ], predicts, a localization length of lattice constants. The measured rates are and, mean, or lattice constants: a discrepancy of. These rates come from evaluating the exact spectral representation ( 62 ) over the ensemble, solving ( 48 ) for each realization and forming ( 53 ); they are distinct from the closed-form average ( 82 ), whose accuracy at this is assessed in section 5.1. Agreement with the established result comes first, the finite- and closed-form advantages afterwards. To make this quantitative, one must evaluate the ensemble average of over the full eigenvalue spectrum, since the logarithm of the transmission amplitude, in place of the amplitude itself, admits a clean spectral representation and a tractable perturbative expansion. This is the content of the exact spectral formulation developed in section 4, where ( 51 ) is recast via Cramer’s rule into a product over eigenvalues, and the ensemble average is evaluated in closed form for both Gaussian and uniform disorder distributions. 4. Exact spectral formulation for localization factor 4.1. Transmission amplitude via Cramer’s rule To derive an exact expression for the transmission amplitude, we return to the matrix equation for harmonic response ( 48 ). For excitation at site with unit amplitude, the force vector is with the nonzero entry at position. The displacement at site can be expressed via Cramer’s rule as where denotes the cofactor of matrix, obtained by deleting row and column from, computing the determinant of the resulting matrix, and multiplying by. For the tridiagonal structure of, the cofactor can be evaluated explicitly. The matrix has the form When row and column are removed, the resulting matrix consists of two uncoupled blocks: one block of dimension in the upper-left, and another block of dimension in the lower-right. For interior sites (away from the boundaries), the determinant of each block factors into products involving the coupling parameter. A detailed calculation shows that for, the cofactor is The key observation is that the cofactor is independent of the specific locations and (apart from the sign factor), depending only on the coupling strength and system size. This remarkable simplification arises from the uniform coupling structure of the chain. The determinant in the denominator of ( 54 ) can be expressed exactly in terms of the eigenvalues. Since is symmetric, it admits the spectral decomposition where are the eigenvalues and are the orthonormal eigenvectors. The determinant is therefore Substituting ( 56 ) and ( 58 ) into ( 54 ), the transmission amplitude from site to site is This exact result expresses the transmission amplitude purely in terms of the eigenvalue spectrum and the coupling parameter. The sign factor reflects the alternating phase of the displacement field along the chain and leaves the amplitude unchanged. The factor represents the baseline transmission determined by the coupling strength: weaker coupling (smaller ) exponentially suppresses transmission. The product in the denominator encodes the resonance structure: transmission is enhanced when the excitation frequency approaches an eigenfrequency, and is suppressed when lies far from all eigenfrequencies. This expression is valid for any disorder realization and any finite system size. It stands free of infinite-system limits, of transfer-matrix propagation and of Lyapunov exponent calculations. The entire transmission behavior is encoded in the discrete eigenvalue spectrum, which can be computed numerically for any specific disorder configuration or treated statistically through perturbation theory as developed in the previous sections. 4.2. Localization factor definition To quantify the spatial decay of wave transmission, we introduce the localization factor, defined as the negative logarithm of the transmission amplitude normalized by the system size: This definition ensures that corresponds to exponential amplitude decay with distance from the source, and indicates lossless transmission. The normalization by makes an intensive quantity, independent of system size in the thermodynamic limit, and directly comparable to the inverse localization length used in infinite-system theories. Substituting the exact transmission amplitude from ( 59 ) into ( 60 ) and taking the absolute value (which eliminates the sign factor ), Using the logarithm product rule,, we obtain the exact spectral representation This result is exact for any disorder realization and any finite, requiring only the eigenvalue spectrum. The first term,, is a geometric factor independent of disorder and excitation frequency: for weak coupling ( ) it is large and positive, indicating strong intrinsic attenuation even for the ordered chain, and for strong coupling ( ) it is negative, facilitating transmission. The second term,, is a spectral average over all eigenvalue detunings from the excitation frequency: modes near resonance contribute large negative values, enhancing transmission, and far-detuned modes contribute positive values, suppressing it. The classical approach to localization derives the localization factor as the Lyapunov exponent of an infinite product of transfer matrices [ 14 ]. The claim in the original version that this approach provides only asymptotic results was too strong and is withdrawn. Finite-size scaling and finite-chain corrections have been developed within the transfer-matrix framework and are reviewed in [ 19 ]. What remains outstanding within it is a closed-form disorder average at fixed finite expressed in the eigenvalue spectrum of the chain. The spectral representation ( 62 ) offers a complementary perspective. Although this representation was noted in the transfer-matrix literature [ 14 ], the dependence of the localization factor on the random eigenvalues as the primary statistical objects remains unexploited, as does the connection to modal phase decoherence. The present work develops this representation systematically. It is exact for any finite without infinite-chain limits, expresses localization directly in terms of the eigenvalue spectrum, separates the disorder-independent baseline from the disorder-induced spectral fluctuations, and permits closed-form disorder averaging through eigenvalue perturbation theory, as demonstrated in the following sections. The spectral sum in ( 62 ) encodes the accumulated phase randomization across modes: disorder perturbs the eigenvalues, causing the logarithmic terms to fluctuate across realizations, and when ensemble-averaged these fluctuations produce the disorder-induced increase in that characterizes Anderson localization. 4.3. Ordered chain baseline Before analyzing disorder effects, we establish the baseline localization behavior of the perfectly ordered chain ( ). Substituting the deterministic eigenvalues from ( 12 ) into the exact spectral formula ( 62 ), where the superscript denotes the ordered-chain value. This expression depends only on the excitation frequency, coupling parameter, and system size. The sum in ( 63 ) evaluates in closed form at finite, which turns the results below into derivations in place of assertions. The ordered dynamic matrix is tridiagonal Toeplitz with diagonal and off-diagonal, so its determinant is a Chebyshev polynomial of the second kind, where is the Bloch wavenumber of ( 6 ) at the excitation frequency. Substituting ( 64 ) into the exact spectral formula ( 62 ) gives the ordered baseline in closed form, Checked against direct evaluation of ( 63 ) to machine precision at for,, and. To understand the frequency dependence, we examine the continuum limit. The discrete sum over modes transitions to an integral by introducing the normalized mode index, so that and the sum becomes In this limit, the deterministic eigenvalues form a continuous band, spanning the passband. The localization factor becomes The behavior depends critically on whether the excitation frequency lies within or outside the passband. For frequencies in the passband,, there exists at least one mode index satisfying, corresponding to exact resonance. Near this resonance, the integrand diverges logarithmically. The cancellation follows at once from ( 65 ). In the passband is real, and is a fixed nonzero constant, so every term in the bracket is bounded independently of and. At, for, and for,. Taking, This result reflects the fundamental property of propagating waves: within the passband, the ordered infinite chain supports Bloch waves with real wavenumbers, and the amplitude decay vanishes. The vanishing localization factor confirms lossless transmission through the perfectly periodic structure. The presence of resonant modes provides channels for energy transfer, and the coherent modal superposition maintains constant amplitude along the chain. For frequencies outside the passband, or, every mode fails to satisfy, and all eigenvalues are detuned from the excitation. The integrand remains bounded, and the integral yields a positive value that exceeds, resulting in In the stopbands, the ordered chain exhibits evanescent decay even without disorder. For (below the passband), the excitation frequency lies below all natural frequencies, and the grounding springs dominate the response, preventing wave propagation. For (above the passband), the excitation frequency exceeds the zone-edge frequency, and the finite coupling ceases to sustain propagating waves. The explicit expression for the stopband localization factor can be derived by evaluating the integral in ( 68 ). For, the result is which diverges as and approaches zero as. For, a similar expression holds: which approaches zero as and increases with further detuning. Both stopband expressions follow the same way. Outside the passband and is complex, equal to below the passband and above it, with; in both branches and. The bracket is then dominated by and, which is ( 71 ) and ( 72 ). At: against for, and against for, the residual being the term. The transition is continuous and carried by, bounded for real and growing as for complex, with at each band edge. Equation ( 65 ) also settles the influence of the fixed boundaries. They enter through and nothing else, this factor being the quantization condition. In the passband the contribution is, bounded above by zero and vanishing as except at the isolated frequencies where is a multiple of and the chain is exactly resonant. In the stopbands it reduces exactly to the bulk rate, so the boundary correction vanishes. Since ( 76 ) acts only on the spectral term of ( 62 ) and leaves untouched, the same holds for the disordered chain: fixed boundaries shift the finite- localization factor by and leave its dependence on, and intact. The band-edge divergence of ( 99 ) is a group-velocity effect, distinct from any boundary effect. These baseline results establish the ordered-chain physics before disorder is introduced. Within the passband, the ordered chain permits perfect transmission ( ), and in the stopbands, transmission is intrinsically forbidden by the band structure. Disorder, as will be shown in subsequent sections, increases the localization factor within the passband, converting propagating waves into localized modes. The disorder-induced increment quantifies the degree of Anderson localization beyond the baseline stopband effects. 5. Weak-disorder perturbation theory and Gaussian averaging The exact spectral formula ( 62 ) holds for any single disorder realization. To obtain statistical predictions relevant for ensembles of disordered chains, we must average over the disorder distribution. The localization factor depends on the eigenvalues through the logarithmic sum, and the eigenvalues in turn depend nonlinearly on the disorder variables even at first order in perturbation theory. This nonlinearity prevents simple term-by-term averaging. However, for Gaussian disorder, the ensemble average can be evaluated exactly in closed form using special functions. For weak disorder strength ( ) and frequencies away from modal resonances, this exact result admits asymptotic expansions that reveal the universal scaling of localization with disorder strength and coupling parameter. This statistical analysis stands free of infinite-system limits: we work directly with the finite- eigenvalue spectrum and take ensemble averages at fixed system size, providing exact predictions for finite chains. 5.1. Exact disorder-averaged localization factor We now derive the exact ensemble-averaged localization factor for Gaussian disorder. Beginning with the spectral formula ( 62 ) and substituting the first-order perturbed eigenvalues from ( 27 ), Define the deterministic detuning of mode from the excitation frequency, so that The first term is deterministic, independent of the disorder realization. The second term fluctuates across the ensemble because the perturbations are random variables. Taking the ensemble average, Define the modal expectation which represents the disorder-averaged logarithmic contribution from mode. The ensemble-averaged localization factor is then To evaluate, we use the fact that for Gaussian disorder with, the perturbation defined in ( 28 ) is also Gaussian with zero mean and variance given by ( 30 ). This is exact at every, in place of asymptotic: is a fixed linear combination of jointly Gaussian variables and that family is closed under linear combination, so the central limit theorem stays out of the argument. Equation ( 82 ) is therefore free of distributional approximation, and its one approximation is the first-order eigenvalue treatment bounded by ( 33 ). The uniform case of section 6.2 differs in exactly this respect. Therefore, is precisely the integral analyzed in appendix B: Applying the result from appendix B, we introduce the dimensionless detuning parameter which measures the ratio of the deterministic detuning to the disorder-induced eigenvalue spread. The exact result is where is Dawson’s integral defined in ( B.12 ) and is Euler’s constant. Substituting ( 81 ) into ( 78 ) yields the exact disorder-averaged localization factor: This result is exact within first-order perturbation theory for the eigenvalues and holds for arbitrary excitation frequency, disorder strength (provided ), and system size. The expression has a clear structure: The term is the geometric baseline from the ordered chain. The term represents the disorder-induced shift in the logarithmic modal weight. The constant arises from the Gaussian averaging and is independent of frequency and disorder strength. The Dawson integral term captures the detuning-dependent modification: modes near resonance ( ) contribute differently than modes far from resonance ( ). The sum over modes in ( 82 ) represents a spectral averaging process. Each mode contributes according to its detuning and its sensitivity to disorder (quantified by ). Modes with small detuning (, corresponding to ) are near resonance and highly sensitive to eigenvalue fluctuations; the Dawson integral for these modes provides the dominant contribution to disorder-induced localization. Modes with large detuning (, corresponding to ) are off-resonant and contribute primarily through the logarithmic terms. To validate the Dawson integral representation ( 82 ), we compare Monte Carlo simulations with the analytical prediction, as shown in figure 6. Figure 6 compares the analytical Dawson integral formulation with Monte Carlo simulations, and the agreement is close in three of the four subfigures. For weak disorder (, subfigures a–b), the analytical expression ( 82 ) captures the localization factor with high precision throughout the passband. The oscillatory structure observed for large coupling (, subfigures (b) and (d)) arises from the same near-resonant sampling mechanism discussed for uniform disorder: when the frequency grid samples close to eigenvalue locations, the integrand in ( 81 ) becomes sensitive to the detailed structure of, producing oscillations that both theory and simulation reproduce accurately. The mechanism is spectral discreteness. Each crossing of an eigenvalue by the excitation frequency changes the sign of one and takes through zero, where the Dawson term in ( 81 ) varies most steeply, giving one extremum of per crossing. Whether crossings are resolved is set by the ratio of the broadening to the mean eigenvalue spacing. At the subfigures of figure 6 have for (b), for (d), for (a) and for (c). Broadened modes separate at small and merge above unity, so the oscillation amplitude falls monotonically along that ordering and vanishes in (c), which is what the figure shows. The oscillation is spectral structure in place of noise: the analytical and Monte Carlo curves oscillate in phase, which independent noise would fail to do. The claim in the original version that agreement is excellent throughout (c) and (d) goes beyond what (c) supports, and is corrected. Agreement is close in (a), (b) and (d). In (c), where is the largest of the four, the analytical curve reproduces the shape and the band-edge divergences but lies below the simulation across the mid-passband, the Monte Carlo values being positive and the analytical values slightly negative. The cause is specific to the localization factor, leaving the eigenvalue statistics intact, which is why figure 3 shows a corresponding success. Equation ( 82 ) perturbs the eigenvalues and retains the ordered eigenvectors, capturing eigenvalue scatter alone, with eigenvector deformation omitted. The eigenvector correction is in the mode shapes and in, the same order as the retained term, with a coefficient growing with. Its onset is set by of ( 83 ), which measures the shift against the level spacing: once broadened eigenvalues overlap, adjacent modes mix and the ordered eigenvectors cease to be an adequate basis. Since at, corresponds to. At, subfigures (a), (b) and (d), the retained term dominates; at in (c), and the two contributions are comparable. An independent check at the same frequency: in the mid-passband and, so ( 99 ) gives for subfigure (c). The Monte Carlo curve there is positive and of that order and the analytical curve is slightly negative, so the closed form is the quantity that departs from the independently derived value, and the simulation agrees with it. The qualitative conclusions are unaffected, each concerning the structure of the average in place of its magnitude. Distribution independence follows from ( 107 ) order by order in, whichever contributions are retained. The identification of follows from ( 95 ), which is exact. The decoherence mechanism is validated at single-mode level in figure 4. What is bounded is the quantitative accuracy of the closed-form localization factor: agreement at the few per cent level for, with shape and band-edge behavior reproduced well beyond it. Numerical evaluation of ( 82 ) is straightforward: compute the deterministic eigenvalues from ( 12 ), evaluate the variances from the exact result ( 32 ), form the dimensionless detunings from ( 80 ), and sum the contributions using a standard implementation of Dawson’s integral. This provides exact predictions for finite chains without requiring Monte Carlo sampling or infinite-system extrapolation. The connection to modal phase decoherence is now quantitative. The Dawson integral terms encode the accumulated phase randomization: modes with experience strong decoherence because the disorder-induced eigenvalue spread is comparable to the detuning, causing large fluctuations in modal amplitudes across realizations. This spectral averaging mechanism is the frequency-domain manifestation of the temporal phase decoherence derived in section 3. 5.2. Weak-disorder expansion The exact result ( 82 ) involving Dawson’s integral provides the complete disorder-averaged localization factor. However, for frequencies well-separated from all modal resonances, an asymptotic expansion yields a simpler closed-form expression that reveals the universal scaling of localization with disorder strength. Consider the off-resonance condition for all modes, which ensures that the dimensionless detuning parameters satisfy. This condition is met when the excitation frequency lies sufficiently far from all eigenfrequencies, so that the deterministic detunings exceed the disorder-induced eigenvalue spread. Under this condition, we expand the logarithm in ( 75 ) as a Taylor series in the small parameter: Using the standard series for, Substituting into ( 85 ), Taking the ensemble average term by term and using the statistical properties of from ( 29 ), where the linear term vanishes because, and the second-order term reduces to the variance. For Gaussian disorder, all odd moments vanish by symmetry:,, etc. The next nonvanishing contribution appears at fourth order, involving the kurtosis. Therefore, to, Substituting into ( 78 ), Recognizing that the first two terms constitute the ordered-chain baseline from ( 63 ), we obtain the weak-disorder result: The disorder-induced correction is second-order in, as expected from perturbation theory. Equation ( 92 ) holds only under the off-resonance condition ( 84 ), where every detuning exceeds the disorder-induced spread. In that regime the correction is negative, so falls below the ordered baseline: away from every eigenfrequency, eigenvalue scatter raises the typical transmission in place of suppressing it, and in a stopband it slightly reduces the band-structure attenuation. Passband localization is the opposite-sign effect and lies outside the reach of this expansion. Within the passband at least one mode satisfies, the ratio diverges, and the localization increment is set by the disorder broadening of these near-resonant modes, captured by the exact Dawson formula ( 82 ) and recovered independently by the continuum result ( 99 ). The physical interpretation connects directly to the phase decoherence mechanism established in section 3. The variance quantifies the eigenvalue scatter for mode, which translates to frequency uncertainty. Modes with small detuning experience large fractional fluctuations in their modal amplitudes and dominate the disorder response, but they are precisely the modes for which the off-resonance expansion fails; their contribution is resolved by the exact Dawson formula ( 82 ), where the diverging logarithm is replaced by a finite disorder-broadened value. The inverse-square dependence on detuning in ( 92 ) reflects the resonant enhancement: modes near the excitation frequency are most sensitive to disorder because their large unperturbed amplitudes magnify the effect of eigenvalue shifts. Passband localization is therefore carried by the broadening of these near-resonant modes, the frequency-domain counterpart of the modal phase decoherence of section 3. Using the asymptotic variance approximation from ( 31 ), valid for large, the disorder correction simplifies to In the continuum limit, the sum transitions to an integral over the density of states, and the disorder correction becomes proportional to with a coefficient determined by the spectral density near the excitation frequency. This scaling,, is universal for weak disorder and has been confirmed extensively in numerical simulations and transfer-matrix analyses [ 14 ]. The weak-disorder expansion ( 92 ) provides a computationally efficient alternative to the exact Dawson integral formula ( 82 ). For frequencies satisfying the off-resonance condition ( 84 ), the asymptotic result is accurate and requires only evaluation of the deterministic eigenvalues and variances. For frequencies near modal resonances, the exact formula must be used to capture the disorder-smoothing of logarithmic singularities. The two descriptions divide the frequency axis at, independently of any feature of the band structure. For the excitation is off resonance from every mode, ( 92 ) applies, and the correction is negative and. For at least one eigenvalue is broadened across the excitation and ( 82 ) is required. Which case holds throughout the passband is decided by of ( 83 ): for, ( 82 ) must be used everywhere, as in figure 6 (c) with; for the regimes alternate across the passband, which is the oscillation of figures 6 (b) and (d). In the stopbands every mode has and ( 92 ) is uniformly valid. 5.3. The continuum limit Connecting the finite- spectral formulation to the classical infinite-chain theory requires taking the continuum limit with care, and provides an independent validation of the scaling identified in the finite-chain analysis. The weak-disorder expansion ( 92 ) loses uniform validity as: for excitation frequencies within the passband, there always exists a mode for which, so the Taylor expansion in powers of breaks down near resonance. The continuum limit must therefore be taken at the level of the governing difference equation, in place of the discrete spectral sum. Starting from ( 48 ), the equation of motion for the -th site of the disordered chain under harmonic excitation is Dividing by and using the ordered dispersion relation ( 6 ), which gives for a wave with Bloch wavenumber and lattice constant, this reduces to the standard one-dimensional tight-binding recurrence with weak diagonal disorder [ 10, 11, 19 ], where and. The disorder-to-coupling ratio is the natural small parameter, consistent with the finite-chain analysis of section 6. To extract the localization factor, we introduce the ratio variable, which satisfies the exact Riccati recursion and the localization factor is the Lyapunov exponent, which is almost surely constant by the ergodic theorem [ 9 ]. For the ordered system the fixed points satisfy, so, giving throughout the passband. For weak disorder we write with, and expand ( 96 ) using and the identity, giving Since, the first-order contribution to vanishes and the leading contribution arises at second order. The denominator in the final result has two distinct origins. The first is the Wronskian of the two Bloch solutions of the ordered recurrence, which equals; by variation of parameters, the amplitude perturbation per step is of order, contributing at second order. The second is the phase average: the correction involves, which averages to over the stationary distribution of the phase, established to be uniform on for generic [ 11, 19 ], introducing an additional factor of in the denominator. Combining these two contributions, the Lyapunov exponent is [ 10, 11 ] Substituting, The localization factor scales with the disorder variance, is suppressed quadratically by the coupling, and diverges near the band edges ( ) where the group velocity vanishes and slow waves scatter most strongly from disorder. It is weakest at the band center ( ), where propagation is fastest. This infinite-chain result is positive throughout the passband and carries the magnitude scaling. It is the continuum counterpart of the disorder broadening of near-resonant modes in the finite-chain formulation, evaluated through the exact Dawson formula ( 82 ), in place of the off-resonance correction ( 92 ), which carries the opposite sign and applies only away from resonance. The two passband descriptions share the same dependence and the same modal phase decoherence mechanism. Three points about the status of ( 99 ). First, it stands independent of ( 82 ), and must: that equation perturbs the eigenvalues and holds the eigenvectors fixed, whereas infinite-chain localization is carried by eigenvector deformation. The two routes are independent and meet in the dependence; the consistency claimed here is that and nothing stronger. Second, the denominator agrees with the established Anderson-model result [ 10, 11, 19 ] and with Pierre [ 14 ], and has been checked numerically in subsection 3.4: against at,,, a discrepancy of. Third, the signs in ( 94 ) to ( 96 ) are corrected relative to the original version. From ( 3 ),