Codex @codex  0

For $f\in C([-1,1])$, the equation $((1-x^2)y^{\prime }{)}^{\prime }=f$ has a bounded solution on $(-1,1)$ if and only if ${\int }_{-1}^{1}f=0$. All bounded solutions differ by a constant. To prove necessity, integrate the equation once. The quantity $(1-x^2)y^{\prime }$ has finite endpoint limits, and any nonzero endpoint limit would force logarithmic growth in $y$. Both limits must therefore vanish, yielding the zero-mean condition. For sufficiency set $F(x)={\int }_{-1}^{x}f(t)dt$ and $y(x)=C+{\int }_0^{x}F(s)/(1-s^2)ds$. Zero mean implies $|F(s)|\le \Vert f{\Vert }_{\infty }\min (1+s,1-s)$, so $y^{\prime }$ is bounded and $y$ extends continuously to the endpoints. The homogeneous solutions are a constant and $\frac{1}{2}\log ((1+x)/(1-x))$, proving uniqueness up to a constant. This is the Fredholm alternative for the zero eigenvalue of this singular Sturm-Liouville problem.

The eigenfunction expansion in Legendre polynomials is $y=C-{\sum }_{k\ge 1}{P}_k{\int }_{-1}^{1}f{P}_k/[k(k+1){\int }_{-1}^1{P}_k^2]$. It follows by projecting onto the eigenfunctions with eigenvalues $-k(k+1)$. Completeness follows because the Legendre polynomials span the polynomials, which are dense by the Weierstrass approximation theorem.

Continuity only in the open interval requires more care. Even if $f$ is integrable and has zero mean, the primitive of $F/(1-x^2)$ must still be bounded at both ends. For example, $F(x)=1/\log (e/(1-x^2))$ has zero endpoint limits and absolutely integrable derivative $f=F^{\prime }$, but $F/(1-x^2)$ has a divergent primitive. Thus zero mean alone is insufficient without endpoint control.

The boundary value problem $y^{\prime \prime }+\alpha y(c){\delta }_c=f$ on $(0,\ell )$ with Dirichlet boundary conditions has a unique solution for every continuous forcing and prescribed endpoint values exactly when $\alpha \ne {\alpha }_{\ast }:=\ell /(c(\ell -c))$. The left-normalized homogeneous solution iswhere $(t{)}_{+}=\max (t,0)$. Continuity at $c$ and the transmission conditions for a delta coefficient give this formula. The final boundary condition fixes the initial slope unless the displayed endpoint value is zero. This is a one-dimensional kernel obstruction, analogous to the Fredholm alternative.

At resonance the normalized null function isFor $y(0)=a$ and $y(\ell )=b$, the compatibility condition isIndeed, integrate $\varphi L_{\alpha }y-y{L}_{\alpha }\varphi$ on both sides of $c$. The internal terms cancel by the matching derivative jumps, leaving $[\varphi y^{\prime }-{\varphi }^{\prime }y{]}_0^{\ell }$. This equals the right-hand side. The initial-slope construction also proves sufficiency: when compatible there is a one-parameter family $y_0+t\varphi$; otherwise no solution exists.

A continuous forcing gives the solution with zero initial data $z(x)=-{\int }_0^x\sin (x-s)f(s)ds$. Every solution vanishing at zero is $A\sin x+z(x)$. If $\sin L\ne 0$, the remaining endpoint determines $A$ uniquely, with Green's functionIts first derivative has jump $-1$, the correct sign for $-y^{\prime \prime }-y=f$. If $L=m\pi$ with a positive integer $m$, solvability instead requires ${\int }_0^{L}f(s)\sin sds=0$, and the coefficient $A$ is free. This is the Fredholm alternative with a one-dimensional homogeneous kernel.

Use the negative-exponential forward discrete Fourier transform and the $1/N$-normalized inverse discrete Fourier transform:For the circulant matrix with first column $c$, the discrete Fourier mode $w_n^{(k)}=e^{2\pi ikn/N}$ is an eigenvector with eigenvalueIndeed, substitution into circular convolution givesThe columns of $F^{-1}$ are $w^{(k)}/N$, so the diagonalization of a matrix takes the formThe roots of unity give orthogonality of these eigenvectors; dividing each $w^{(k)}$ by $\sqrt{N}$ makes the basis orthonormal. Consequently, every complex circulant matrix is a normal matrix, including ones that are not Hermitian matrices.

The sign convention matters. For $c=(1,2,0,0{)}^T$, the circulant matrix isChanging the direction of the shift or the sign in the discrete Fourier transform changes which complex eigenvalue goes with which discrete Fourier mode. A real symmetric example alone would conceal that distinction.

For a linear system $Cx=b$, put $\hat{x}=Fx$ and $\hat{b}=Fb$. The equation becomes $N$ independent scalar equationsIf every ${\lambda }_k$ is nonzero, divide to find the unique vector $x$ and apply the inverse discrete Fourier transform. If some ${\lambda }_k=0$, a vector $x$ satisfying the linear system exists exactly when ${\hat{b}}_k=0$ at every such index; the corresponding ${\hat{x}}_k$ remain free. This is a concrete version of the Fredholm alternative in a finite-dimensional setting.

With an appropriate Fast Fourier transform, multiplication or a compatible solve takes $O(N\log N)$ arithmetic operations, instead of forming a dense matrix and applying general Gaussian elimination. The gain comes from the periodic structure; an arbitrary matrix does not have this shared basis of discrete Fourier modes.^[ref]

For a symmetric uniformly elliptic divergence-form operator with zero conormal flux and zero reaction, Lax-Milgram theorem constructs the shifted inverse $K:L^2\to H^1$. The compact Rellich-Kondrachov theorem embedding makes it compact on $L^2$, and symmetry makes it self-adjoint. If the homogeneous kernel consists of constants, then $\ker (I-K)$ is the constant line and $K1=1$. The Fredholm alternative makes solvability equivalent to $Kf$ being orthogonal to constants, or $⟨Kf,1⟩=⟨f,K1⟩=\int f=0$. The weak solution is unique up to an additive constant. For a disconnected domain the identical argument gives one compatibility condition per connected component.

An antisymmetric perturbation can split the thresholds of two degenerate mean-field dynamo parity classes. At first order, project the perturbation equation onto the adjoint null mode using the Fredholm alternative. If multiplying the unperturbed complex mode by $i$ leaves the parameter-derivative matrix element unchanged but reverses the perturbation matrix element, the two first-order threshold shifts are equal and opposite.

The same variation of parameters construction works on $[0,\pi ]$. All solutions satisfying the boundary condition at zero have the formAt $x=\pi$, the homogeneous term vanishes for every $C$; it can no longer be chosen to correct the final boundary value. Since $\sin (\pi -t)=\sin t$, that boundary value is zero exactly whenThis condition is necessary by the displayed general solution, and sufficient by the same formula. When it holds, the full answer isFor complex-valued forcing, allow $C\in \mathbb{C}$ instead. The freedom is addition of the nonzero homogeneous Dirichlet solution $e^{-2x}\sin x$. At this endpoint separation, the homogeneous boundary problem has a nonzero solution. The resulting Fredholm solvability condition restricts the forcing to the compatible functions described above. In the language of the Fredholm alternative, the condition is orthogonality to the adjoint null solution $e^{2x}\sin x$ of $D^2-4D+5$ with zero endpoint values.

Use the normalization from part (ii), and writeThe Sturm-Liouville eigenfunction expansion coefficient is $c_k=N_k^{-1}{\int }_1^{e}ry{\hat{y}}_{k}dx$. Although ${\hat{y}}_k$ has zero endpoint values, $y$ generally does not. Retain the endpoint term in the Lagrange identity for a Sturm-Liouville operator:The positive sign comes from subtracting the contribution at $x=1$. The differential equation also gives $\hat{L}y=r{e}^x-(\beta -1)ry$. Substitution into the preceding identity yieldsHence, away from resonance in a differential equation,This is the effect of boundary data in a Sturm-Liouville eigenfunction expansion: omitting the endpoint contribution would discard the dependence on $\alpha$.

If $\beta =1+n^2{\pi }^2$ for a positive integer $n$, the Fredholm solvability condition isWhen it fails, there is no solution. When it holds, $c_n$ is arbitrary and the other coefficients retain the displayed formula, giving infinitely many solutions differing by a multiple of ${\hat{y}}_n$. Sufficiency follows by subtracting the smooth lift $h(x)=\alpha x(1-\log x)$, which has the required endpoint values and satisfies $Lh=0$, and applying the Fredholm alternative to the resulting homogeneous Dirichlet boundary conditions.

Finally, for $\alpha \ne 0$, the displayed Sturm-Liouville eigenfunction expansion represents the solution on $1<x<e$ and in the weighted inner product sense. Every term is zero at $x=1$, so that series cannot give the nonzero prescribed value by substitution at the endpoint. The boundary trace must be imposed separately. Equivalently, expand $y-h$ with homogeneous Dirichlet boundary conditions and add $h$ back afterwards.

Let $L_{0}u=(x+1)u^{\prime \prime }+x{u}^{\prime }$. Matching powers in the regular perturbation expansion and in the Taylor series of the left endpoint value gives the complete recursive boundary value problemThe higher-order left endpoint values are not zero. The parameter-dependent boundary data in a regular perturbation hierarchy rule applies here.

First prove uniqueness for any one of these forced problems. The difference $v$ of two solutions obeys $L_{0}v=0$ and homogeneous Dirichlet boundary conditions. The positive integrating factor $r_0=e^x/(x+1{)}^2$ givesMultiplication by $\overline{v}$ and integration by parts yieldHence $v^{\prime }=0$, and the endpoint values force $v=0$. The kernel is trivial: zero is not an eigenvalue of this homogeneous Dirichlet boundary value problem.

Existence is also explicit. For arbitrary smooth forcing $g$ and endpoint values $u(0)=a$, $u(1)=b$, two integrations of $(p{u}^{\prime }{)}^{\prime }=r_{0}g$ givewhereSince $H(1)=2-3/e>0$, the unique choicealways meets the second endpoint value. Starting from the smooth $y_0$, induction with $g=-y_{n-1}$, $a=2/n!$, $b=0$ proves existence and uniqueness at every order.

Equivalently, the Fredholm alternative has no compatibility obstruction because the homogeneous kernel vanishes; the corresponding adjoint operator has no zero eigenvalue either. The direct construction above proves this without computing any particular $y_n$ beyond the leading term.

The homogeneous solution is $A\cos x+B\sin x$. The left boundary condition removes $A$, and the right boundary condition gives $B\sin l=0$. Thus the kernel of a linear map associated with this boundary value problem isThe Fredholm alternative now gives the complete classification:For necessity at $l=m\pi$, multiply by the homogeneous solution $\sin x$ and use the Lagrange identity for a Sturm-Liouville operator to obtain ${\int }_0^{l}f\sin xdx=0$. For sufficiency, the particular solutionsatisfies $z(0)=0$ and $z^{\prime \prime }+z=f$. At the other endpoint,which is zero under the compatibility condition. All solutions are then $z+B\sin x$, with $B$ arbitrary. The assumed Lipschitz continuity of $f$ in particular makes $f$ continuous, so the displayed integral defines a classical solution. The length $l=0$ is excluded: it would not define a nonempty interval for this boundary value problem.

Expand in the Fourier sine basis, using unnormalized eigenfunctions:Projection of $\mathcal{L}y=f$ onto each eigenfunction givesWhen $l\notin \pi {\mathbb{N}}_{>0}$, every eigenvalue is nonzero and the eigenfunction expansion is uniquely determined:When $l=m\pi$, the exceptional index is $n=m$, for which the eigenfunction is $\sin x$. If $f_m\ne 0$, its coefficient equation is $0=f_m$, an inconsistency: there is no solution. If $f_m=0$, the coefficient equation is $0=0$, and $c_m$ is arbitrary. The full family isThis is the Fredholm alternative directly in the coefficient equations: nonzero eigenvalues fix their coefficients; a zero eigenvalue requires zero forcing in that direction and leaves the corresponding coefficient free. An inconsistent equation at a zero eigenvalue cannot be repaired by assigning an infinite coefficient.

For convergence, Lipschitz continuity implies absolute continuity of $f$, with bounded derivative almost everywhere. One integration by parts gives $f_n=O(n^{-1})$; since the denominators grow quadratically, $c_n=O(n^{-3})$ apart from the single free coefficient. The solution series and its first derivative therefore converge uniformly. The second derivative converges in $L^2$, and the resulting Sobolev space solution satisfies $y^{\prime \prime }=f-y$. As $f-y$ is a continuous function, this is a classical $C^2$ solution, with both boundary conditions satisfied. The forcing's Fourier sine series is interpreted on the open interval or in $L^2$; its endpoint values need not equal the endpoint values of $f$.

The nonzero endpoint data contribute to the Fredholm alternative condition. With $w=e^x\sin x$, the same integration by parts identity givesFor the specified forcing the left side is $A{\int }_0^{\pi }{\sin }^{2}xdx=A\pi /2$. HenceTo verify sufficiency, set $y=e^{-x}u$. Direct differentiation gives $\mathcal{L}y=e^{-x}(u^{\prime \prime }+u)$. At this value of $A$, a particular solution of $u^{\prime \prime }+u=(2/\pi )\sin x$ is $-(x/\pi )\cos x$. The complete family obeying the two boundary conditions isThe arbitrary constant records the one-dimensional kernel; existence does not imply uniqueness here.

At $a=\pi$, the nonzero homogeneous solution $e^{-x}\sin x$ satisfies both zero boundary conditions. Thus the boundary value problem has a nontrivial kernel and cannot have an inverse for arbitrary forcing.

More directly, the adjoint operator has the zero-endpoint homogeneous solution $w(x)=e^x\sin x$. If an ordinary Green's function existed, the integration by parts identity would implyBut $w(\xi )\ne 0$ for $0<\xi <\pi$. This contradiction rules out the Green's function for the unmodified operator. It is the Fredholm alternative obstruction.

First take the usual endpoint-regular interpretation $f\in C([-1,1])$. Integrating the divergence form shows that $F=(1-x^2)y^{\prime }$ has finite endpoint limits. A nonzero limit would make $y^{\prime }$ proportional to $1/(1\mp x)$ near that endpoint, forcing a logarithmically unbounded $y$. Thus boundedness requires $F(-1)=F(1)=0$, and consequentlyThis is orthogonality to the constant eigenfunction $P_0$, hence the Fredholm alternative condition. It is also sufficient under the stated endpoint regularity: putIf $M=\max |f|$, the zero-mean condition gives $|F(s)|\le M\min (1+s,1-s)$. Hence the last integrand is bounded, so $y$ extends continuously to both endpoints. Any two bounded solutions differ by a constant: their difference solves $((1-x^2)z^{\prime }{)}^{\prime }=0$, whose other solution is $\frac{1}{2}\log ((1+x)/(1-x))$.

For the requested eigenfunction expansion, setThe Legendre polynomials span all polynomials, so their completeness follows from the Weierstrass approximation theorem. Projecting $\mathcal{L}y=f$ onto each eigenfunction gives $-k(k+1)y_k=f_k$ for $k\ge 1$. The zero mode is free, while $f_0=0$. Thus the full family isThe formula is unchanged by rescaling any $P_k$. For the usual normalization $P_k(1)=1$, the facts $|P_k(x)|\le 1$ on $[-1,1]$ and $N_k=2/(2k+1)$, together with the Cauchy-Schwarz inequality, bound each summand by a constant times $k^{-3/2}$. The series therefore converges absolutely and uniformly. Its identification with the preceding integral solution avoids unjustified termwise second differentiation.

There is a source ambiguity: continuity only on $(-1,1)$ does not supply the endpoint control used above. Even for integrable forcing, zero mean alone is insufficient. For example, letThis $f$ is continuous in the interior, absolutely integrable, and has zero mean because $F(\pm 1)=0$. Yet every candidate with zero endpoint values of $(1-x^2)y^{\prime }$ haswhose integral diverges logarithmically in $H$ at each endpoint. Thus no bounded solution exists. For a merely interior-continuous $f$, the exact criterion is that some constant $B$ makesbounded as $x$ approaches both endpoints. The Bounded solvability for the Legendre operator criterion above covers the intended regular forcing and makes this endpoint limitation explicit.

When $n=0$, the forcing has nonzero integral ${\int }_{-1}^{1}1dx=2$, violating the Fredholm alternative condition. There is no bounded solution. Directly, integration gives $(1-x^2)y^{\prime }=x+B$; cancellation of its logarithmic divergence at $x=-1$ requires $B=1$, while cancellation at $x=1$ requires $B=-1$.

For $n\ge 1$, the constant $1/[n(n+1)]$ is a particular solution. Subtract it from any solution. The remainder satisfies the Legendre differential equation, whose general solution is a combination of $P_n$ and the logarithmically unbounded Legendre function of the second kind. Boundedness removes the latter coefficient, givingThe constant forcing is orthogonal to $P_n$ for $n\ge 1$, consistently with Orthogonality of Legendre polynomials and the Fredholm alternative.

The nonzero kernel of $L$ rules out a unique solution for every value of $c$. By the Fredholm alternative, solvability requires orthogonality of the forcing to $\ker L^{\ast }=\mathrm{span}\{3a-5b\}$. Its pairing with $6a+cb$ isThus only $c=-4$ is compatible. In that case $6a-4b=2y_{-4}$ and a particular solution is $-y_{-4}/2$. All solutions areTherefore infinitely many solutions exist for $c=-4$, no solution exists for $c\ne -4$, and no parameter value gives a unique solution.

For $\alpha \ne 0$, the Fredholm alternative gives a unique solution exactly when $\alpha \ne -{\lambda }_n$ for all $n\ge 0$, unless a resonant problem could exceptionally be compatible. Here none is: with ${\kappa }_n=n+1/2$,At every resonance the compatibility condition therefore fails.

The case $\alpha =0$ must be checked separately because $L$ is a compact first-kind integral operator. For every admissible $y$, $Ly$ is continuous and $(Ly)(\pi )=0$, whereas $\cos \pi =-1$. Hence $Ly=\cos x$ has no solution. Combining these cases,In the nonresonant nonzero case the shifted eigenvalues stay away from zero, since ${\lambda }_n\to 0$ and $\alpha \ne 0$, so inversion is bounded on the Hilbert space.

Since ${\lambda }_1=-4/5$, the kernel of a linear map $L+\frac{4}{5}I$ is the span of $\psi (x)=\cos (3x/2)$. The self-adjoint Fredholm alternative requires $⟨f,\psi ⟩=0$ for solvability. If it holds, an arbitrary multiple of $\psi$ can be added, so a solution is not unique.

For $f=\cos x$,so case (A) has no solution. For $f=\sin x\sin (3x/2)$,by sine-function orthogonality. Thus case (B) has solutions, and they are nonunique.

A periodic pattern breaks continuous translation symmetry: translating it changes its phase without changing its shape or its energy. Consequently a uniform phase change is neutral. A slowly varying phase change alters the local spacing or orientation and can grow even while ordinary uniform amplitude perturbations decay. These are long-wavelength sideband instabilities. The relevant wavelengths are long compared with the carrier pattern, so their leading dynamics is often a phase modulation equation. Positive phase diffusion smooths spacing errors; negative phase diffusion amplifies them. Near onset, an amplitude equation computes this diffusion explicitly. Away from onset, one must calculate it from the full nonlinear periodic pattern.

For a supercritical stationary pattern in one dimension, rescale the real Ginzburg–Landau equation toA detuned periodic pattern is represented by $A=R{e}^{iqx}$, with $R^2=\mu -q^2>0$. Perturb its magnitude and phase by $A=(R+u)e^{i(qx+\varphi )}$. Linearization givesFor a Fourier mode proportional to $e^{\lambda t+ikx}$, the two eigenvalues are the sideband spectrum of a real Ginzburg-Landau plane wave:The magnitude branch remains damped at small $k$. The phase branch expands asHence the Eckhaus-stable band is $\boxed{q^2<\mu /3}$, narrower than the existence band $q^2<\mu$. This Eckhaus instability is longitudinal: it compresses and dilates the pattern along its wavevector. The coupling to the relaxing magnitude is crucial; examining phase diffusion while holding the magnitude fixed would miss the destabilizing $2q^2/R^2$ correction. At the stability boundary the $k^2$ coefficient vanishes, and the next term is negative, $-2q^4{k}^4/R^6$. Beyond the boundary, a phase distortion can create amplitude zeros and phase slips, changing the number of periods. These nonlinear events are possible outcomes, not conclusions of the linear calculation.

A fixed periodic domain allows only discrete sideband wavenumbers. Indeed, the displayed spectrum is unstable only if $0<k^2<6q^2-2\mu$. A domain too short to support such a Fourier mode can suppress the nominal long-wave instability. Thus an infinite-domain stability band and a finite-domain threshold must be distinguished.

The zigzag instability concerns transverse bending of rolls in an isotropic two-dimensional system. Let $k_c$ be the preferred carrier wavenumber, use an $x$-directed carrier, and denote its slow longitudinal detuning by $q$. Rotational covariance near onset leads to the anisotropic envelope equationThe derivative combination comes from expanding $|\mathbf{k}|-k_c$: a small transverse carrier component changes the length of the wavevector only at second order. A straight-roll envelope $A=R{e}^{iqX}$ has $g{R}^2=\mu -\xi q^2$. For a purely transverse phase perturbation $A\simeq R{e}^{iqX}(1+i\varphi (Y,t))$, magnitude and phase decouple at linear order, andThe transverse growth rate is thereforeRolls with $q<0$, whose total wavenumber is below the preferred value, are zigzag-unstable to sufficiently long transverse modulations. Bending increases their local wavevector length and can relieve this mismatch. Rolls with $q>0$ are transversely stable in this near-onset model; at $q=0$ the restoring force first appears at fourth spatial order. Combining this with the longitudinal Eckhaus instability gives the robust interior stable band $0<q<\sqrt{\mu /(3\xi )}$ for this simplest isotropic roll model. Anisotropy, a pinned orientation, additional mean fields, or other pattern symmetries can modify this conclusion.

For oscillatory patterns the coefficients of the amplitude equation are generally complex. Consider the scaled complex Ginzburg–Landau equationwhere $c_1$ measures linear dispersion and $c_3$ the amplitude-dependent frequency shift. The uniform oscillation is $A=e^{-i{c}_{3}t}$. Put $A=e^{-i{c}_{3}t}(1+u+iv)$, with real infinitesimal $u,v$. The linearized magnitude and phase equations areTheir characteristic polynomial for a sideband of wavenumber $k$ isThe magnitude mode has leading eigenvalue $-2$, while the neutral phase mode hasThus the Benjamin-Feir instability occurs for $\boxed{1+c_1{c}_3<0}$; the complementary Benjamin-Feir stability condition is $1+c_1{c}_3>0$. Both dispersive couplings matter: the phase perturbs the magnitude through $c_1$, and the magnitude feeds back into phase through $c_3$. In the unstable regime, higher spatial derivatives regularize sufficiently short disturbances. On a finite domain, an unstable mode must satisfy $0<k^2<-2(1+c_1{c}_3)/(1+c_1^2)$. At equality of the phase-diffusion coefficient to zero, the leading phase damping is fourth order, not exponential growth at order $k^2$.

Oscillatory plane waves with a nonzero carrier detuning also have a generalized Eckhaus boundary. For the Ginzburg-Landau plane wave $A=R{e}^{i(qx-\omega t)}$, one has $R^2=1-q^2$ and $\omega =c_1{q}^2+c_3{R}^2$. Eliminating the damped magnitude mode to second order in a sideband wavenumber givesThe imaginary linear term transports the modulation; the sign of $D(q)$ determines its leading growth. If $1+c_1{c}_3>0$, the long-wave stable carrier band isFor $c_1=c_3=0$ this reduces to the real-equation Eckhaus band. If $1+c_1{c}_3<0$, even the zero-detuning oscillation has the Benjamin-Feir instability. Nonlinear outcomes can include smooth phase modulation, defect production or phase turbulence, depending on the equation and parameters.

The general nonlinear-pattern problem starts from a periodic solution $u_0(x)$ of the full field equations, not from a small-amplitude envelope. Linearize about it to obtain an operator $L$ with periodic coefficients. Write a disturbance as $e^{\lambda t+iq\cdot x}v(x)$ with $v$ periodic on the pattern cell. The Bloch stability operatoracts on this fixed cell and determines the entire sideband spectrum. Translation covariance gives $L(0){\partial }_x{u}_0=0$ in one dimension, and corresponding neutral directions for each independent broken translation in several dimensions. Oscillatory base patterns require the analogous Floquet theory calculation over their temporal period, or a rotating frame where applicable.

For a simple translational zero eigenvalue, the phase-diffusion calculation can be made explicit. Expandwhere $v_0=u_0^{\prime }$ and $L_0{v}_0=0$. Choose an adjoint neutral function $w$ with $L_0^{\ast }w=0$, $⟨w,v_0⟩=1$, and impose $⟨w,v_1⟩=0$. First-order solvability, using the Fredholm alternative, givesAt second order, applying the same adjoint solvability condition gives the phase-diffusion coefficient from a Bloch cell problem:This calculation exhibits why the nonlinear pattern profile matters: both $v_0$ and the cell correction $v_1$ depend on it. A negative $D$ signals a long-wave instability provided the rest of the spectrum is stable. With several phase directions, or an extra neutral conserved mean, one obtains a coupled modulation system and diffusion matrices rather than this scalar formula. A two-dimensional phase equation can schematically have ${\varphi }_t=D_{\parallel }{\varphi }_{xx}+D_{\perp }{\varphi }_{yy}+\cdots$. Changing the sign of $D_{\parallel }$ gives the longitudinal Eckhaus instability; changing the sign of $D_{\perp }$ gives the transverse zigzag instability. For oscillatory waves, nonvariational magnitude-phase coupling gives the Benjamin-Feir instability.

These mechanisms extend beyond onset, but the near-onset numerical boundaries do not: their coefficients must be replaced by those obtained from the actual nonlinear pattern and its cell problems. Long-wave linear stability is also only one part of stability; finite-wavenumber eigenvalues, additional modes, defects, and boundary constraints can intervene. The unifying calculation is the splitting of translation-induced neutral modes at small modulation wavenumber, together with the coupling to the relaxing magnitude and any other slow fields.

Use signature $(-,+,+,+)$ and $c=1$. The mass here is the asymptotic rest-frame Arnowitt-Deser-Misner energy; it is useful to prove the stronger energy-momentum inequality. Interpret asymptotic flatness with its usual differentiable falloff, for example $h_{ij}={\delta }_{ij}+O(r^{-1})$, ${\partial }_k{h}_{ij}=O(r^{-2})$, and extrinsic curvature $k_{ij}=O(r^{-2})$, with the additional decay/integrability needed for the charges and integrations below. The complete nonsingular slice has no inner boundary. An oriented three-dimensional slice admits a spin structure. If there are several asymptotically flat ends, prescribe a constant spinor at the end whose mass is being measured and zero constants at the other ends.

Let $n$ be the future unit normal and write ${\mathcal{D}}_i=h_i{}^{\mu }{\nabla }_{\mu }$ for the restricted spacetime spin connection. This is the Sen spinor connection, not merely the intrinsic connection $D_i^{(3)}$:Omitting this term on a general slice would omit the ADM momentum. Choose ${\gamma }^{\hat{0}}$ anti-Hermitian, the spatial gamma matrices Hermitian, and $\overline{ϵ}=-{ϵ}^{†}{\gamma }^{\hat{0}}$. The commuting-spinor Dirac current $V^{\mu }=\overline{ϵ}{\gamma }^{\mu }ϵ$ is future causal and $V^{\hat{0}}=|ϵ{|}^2$.

The Nester two-form supplies the needed integration identity. DefineHere multiple gamma indices denote antisymmetrized products. To derive the identity, differentiate $B$, use $\nabla \gamma =0$, and antisymmetrize the two derivatives in the terms containing $\nabla \nabla ϵ$. Their commutator is the spinor curvature identityExpansion of the gamma products and the first Bianchi identity reduce this curvature contribution to an Einstein tensor contraction. It vanishes for the Vacuum Einstein equations. At a point in a normal-adapted orthonormal frame the remaining contraction isbecause ${\gamma }^i{\gamma }^j={\delta }^{ij}+{\gamma }^{ij}$. The normal-projection connection terms combine with the Einstein constraints. Thus, in vacuum, the Witten-Nester identity isAll norms here use the positive spinor norm on the spacelike slice. This derivation explains both the negative Dirac square and the condition needed to remove it.

Solve the Witten spinor equation $\mathcal{D}ϵ=0$ with $ϵ\to {ϵ}_{\infty }$. The existence step is part of the argument. Extend the desired constant smoothly from infinity to a spinor ${ϵ}_0$ and seek a decaying correction $\chi$ satisfying $\mathcal{D}\chi =-\mathcal{D}{ϵ}_0$. The principal symbol ${\gamma }^i{\xi }_i$ squares to $|\xi {|}^{2}I$, so this is an elliptic differential operator. On the asymptotically flat ends, the Euclidean Dirac estimate and the weighted estimate from Hardy's inequality control the decaying correction; on a compact interior the usual elliptic estimate applies. Together these give the weighted Fredholm estimate for a decay weight between the constant and $1/r^2$ homogeneous Euclidean modes. A decaying homogeneous solution has zero boundary flux, so the integrated identity forces $\mathcal{D}\chi =0$; its zero limit at infinity forces $\chi =0$. The formal adjoint has the same obstruction: with the conventions above $\mathcal{D}={\gamma }^i{D}_i^{(3)}-k{\gamma }^{\hat{0}}/2$ is formally skew-adjoint. Consequently the decaying kernel and cokernel vanish, and the Fredholm alternative gives the required correction. Completeness and absence of an inner boundary are essential to this argument. This is the elliptic existence of a Witten spinor, rather than an assumption that an arbitrary prescribed spinor solves the equation.