Singular perturbation ($ϵ\to 0^{+}$)

A singular perturbation is a parameter-dependent problem whose limiting problem changes qualitatively as the parameter tends to zero. In an ordinary differential equation, a small coefficient of the highest derivative can reduce the order of the limiting differential equation, preventing it from satisfying all the original boundary conditions. Different length or time scales often require different asymptotic expansions.^[ref]

Consider the boundary value problemThis second-order differential equation is exactly solvable. Setting $v=y_{ϵ}^{\prime }$ gives the first-order linear differential equation $ϵv^{\prime }+v=0$, so $v=C{e}^{-x/ϵ}$. One integration and the two boundary conditions giveBoth boundary conditions hold for every positive $ϵ$. For numerical evaluation, write the same ratio as $expm1(-x/ϵ)/expm1(-1/ϵ)$, where $expm1(t)$ evaluates the exponential function minus one without catastrophic cancellation near $t=0$. Keep the denominator for finite $ϵ$: at $ϵ=0.2$ it is approximately $0.993262$, rather than exactly one.

Setting $ϵ=0$ gives $y_0^{\prime }=0$. The outer solution selected by the right boundary condition is therefore $y_0=1$, which fails the left boundary condition. The exact exponential function $e^{-x/ϵ}$ shows how that mismatch is repaired: the solution rises from zero to almost one within a boundary layer of width $x=O(ϵ)$, using Big O notation. For each fixed $x>0$, $y_{ϵ}(x)\to 1$, while $y_{ϵ}(0)=0$ always. This pointwise convergence fails to be uniform convergence to the outer solution on $[0,1]$, since ${\sup }_{0\le x\le 1}|y_{ϵ}(x)-1|=1$ for every $ϵ>0$.

The coefficient $ϵ$ is small, but its product with the second derivative need not be small. Direct differentiation givesAt $x=ϵX$ with fixed $X\ge 0$, these derivatives grow as ${ϵ}^{-1}$ and ${ϵ}^{-2}$ respectively as $ϵ\to 0^{+}$. Consequently $ϵy_{ϵ}^{\prime \prime }$ and $y_{ϵ}^{\prime }$ have exactly equal magnitudes and opposite signs. Even away from the boundary layer, their magnitudes remain equal in this example; both become exponentially small at each fixed positive $x$. The outer expansion neglects an exponentially small transient there, rather than establishing that one derivative term is smaller than the other everywhere.

To predict the width without knowing the exact solution, suppose an order-one change occurs over distance $\delta$. The derivative scales are $y^{\prime }=O({\delta }^{-1})$ and $y^{\prime \prime }=O({\delta }^{-2})$. The dominant balance $ϵ/{\delta }^2\sim 1/\delta$ gives $\delta \sim ϵ$. The resulting inner variable $X=x/ϵ$ makes the boundary layer visible at a fixed scale. This is the useful feature of singular perturbation: an apparently disappearing term identifies the missing structure when one measures its derivatives on the appropriate scale.

For $\ddot{x}+k(x^2-1)\dot{x}+x=\lambda$, with fixed $|\lambda |<1$ and $k\to \infty$, the Liénard coordinate $y=\dot{x}/k+x^3/3-x$ has fast horizontal jumps and slow motion on $y=x^3/3-x$. The slow time is $t/k$. Integrating $(x^2-1)dx/dT=\lambda -x$ along $[-2,-1]$ and $[2,1]$ gives the displayed period. The two jumps take negligible time at this scale. For fixed $\lambda >1$, the right attracting branch contains the stable equilibrium and the large relaxation cycle disappears.

For a fast-slow system $ϵ\dot{x}=F(x,y)$, $\dot{y}=G(x,y)$ with $0<ϵ\ll 1$, the rescaled time $\tau =t/ϵ$ gives $x_{\tau }=F$, $y_{\tau }=ϵG$. Away from zeros of $F$, an order-one change of $x$ takes order-one fast time but only order-$ϵ$ original time, while $y$ changes little. Setting $ϵ=0$ in these rescaled equations freezes $y$ and produces the leading fast subsystem. This limit differs from the slow reduction $F=0$ in the original time; matching the two describes a relaxation oscillation.

For a fast-slow system $ϵ\dot{x}=f(x,y)$, $\dot{y}=g(x,y)$, the formal critical manifold is $f=0$. Where the fast linearization has eigenvalues with strictly negative real parts, trajectories rapidly approach a nearby attracting slow manifold and then evolve on the slow time. Near folds or loss of this spectral condition, simply imposing $f=0$ may fail.

An initial layer is a short transient needed when the reduced outer solution cannot satisfy the original initial condition. On an inner time scale, terms multiplied by a small parameter remain comparable to the retained dynamics. For $ϵc_t=1-(1+\beta )c$, $c(0)=0$, the inner solution $c=(1-e^{-(1+\beta )t/ϵ})/(1+\beta )$ matches the outer value $1/(1+\beta )$ while retaining the initial datum.

For $ϵ>0$, the two terms in $y=x^{-1/ϵ}+x$ are equal at $x=1$. Their ratio is larger than one for $0<x<1$ and smaller than one for $x>1$. As $ϵ\to 0^{+}$ it tends to infinity at fixed $x<1$ and to zero at fixed $x>1$. Comparable contributions require $|\log x|=O(ϵ)$, equivalently $x=1+O(ϵ)$.

To resolve this region exactly, introduce the inner variable $X=\log x/ϵ$. Then $y=e^{-X}+e^{ϵX}\sim e^{-X}+1$ for bounded $X$. At fixed $x>1$, the first term $\exp [-(\log x)/ϵ]$ is smaller than every positive integer power of $ϵ$, so a regular perturbation in powers of $ϵ$ misses it. At fixed $x<1$, that same term diverges. This explains why a power series approximation valid on one side need not be a uniform approximation across the crossover.

For a system $\epsilon \dot{x}=F(x,y)$, $\dot{y}=G(x,y)$ with fast variable $x$ and slow variable $y$, the critical manifold is the set of equilibria of the fast problem $dx/ds=F(x,y)$ with $y$ frozen. It is attracting where the fast Jacobian matrix has eigenvalues with negative real parts and repelling where those real parts are positive. A smooth graph can cease to be attracting at a fold point of a critical manifold.

For one fast and one slow variable, a nondegenerate fold point of a critical manifold satisfies $F=F_x=0$ and $F_{xx}{F}_y\ne 0$. An attracting and a repelling branch of fast equilibria meet there. A slow trajectory reaching the fold generally leaves by a rapid jump when its slow drift crosses the fold transversely. The fold is defined relative to the fast direction: if fast trajectories are slanted in the plotted coordinates, their tangency to the critical manifold need not be an extremum of its plotted height.

Near a nondegenerate fold point of a critical manifold crossed with nonzero slow speed, the local balance can be written $\epsilon \dot{x}=y-x^2$, $\dot{y}=-1$ up to nonzero scalings and higher-order terms. A time interval $T$ gives $y=O(T)$; balancing $x^2$ with $y$ gives $x=O(T^{1/2})$, and balancing $\epsilon x/T$ with $T$ gives $T=O({\epsilon }^{2/3})$, $x=O({\epsilon }^{1/3})$. This layer delays departure from the fold relative to an instantaneous singular jump. It differs from the order-$\epsilon$ fast motion away from the fold.

A singular limit is a parameter limit in which the governing problem loses an order, changes type, or otherwise changes its qualitative solution structure. For $\epsilon \dot{x}=F(x,y)$, setting $\epsilon =0$ produces the algebraic constraint $F(x,y)=0$ and can omit fast transients required by the initial data.

In the rigid-backed shear-wave dispersion equation $\tan \theta =-ϵ\theta /\sqrt{A^2-{\theta }^2}$, take fixed $(n-1/2)\pi <A<n\pi$ and let $ϵ\to 0^{+}$. The highest intersection approaches the endpoint $A$. Setting $\delta =A-\theta$ and using $A^2-{\theta }^2\sim 2A\delta$ gives the displayed asymptotic expansion. At $A=n\pi$ the appropriate balance is instead $\delta \sim ({ϵ}^{2}n\pi /2{)}^{1/3}$. Near that transition, with $\Delta =A-n\pi$, the leading equation is ${\delta }^2(\Delta +\delta )={ϵ}^{2}n\pi /2$, where now $\theta =n\pi -\delta$ and $\Delta +\delta >0$.

For a weakly perturbed harmonic oscillator, expand both its periodic solution and its angular frequency, using the stretched time $\tau =\omega t$. At each order, choose the frequency correction to remove fundamental-harmonic forcing that would produce a secular term. For a self-excited periodic orbit, cancellation of the independent sine and cosine terms can determine both the leading amplitude and the frequency correction. Actual existence follows when the associated Poincaré return map has a simple displacement zero, by the implicit function theorem.

A homogeneous mode can be order one inside an interval and exponentially small at an endpoint. Adjusting an endpoint value beyond all algebraic orders can then change the interior solution by order one. In the nested layers at an essential singular endpoint example, the mode $\exp (-x/ϵ-2{ϵ}^2/x)$ has this behavior: prescribed exact endpoint values give a unique solution for each fixed positive $ϵ$, but leading asymptotic endpoint data do not select a unique bounded family.

Adiabatic elimination replaces a rapidly relaxing variable by its instantaneous stable quasistatic value. If $\dot{R}=-\Gamma (t)R+P(t)$ with $\Gamma >0$, write $R=R_0+\delta R$, $R_0=P/\Gamma$. Then $\dot{\delta R}+\Gamma \delta R=-{\dot{R}}_0$. After the transient, $\delta R\simeq -{\dot{R}}_0/\Gamma$ when coefficients vary slowly compared with ${\Gamma }^{-1}$. This gives both the leading reduction and its time-scale error; rapid relaxation alone does not justify an additional Taylor expansion in another parameter.

A matched asymptotic expansion constructs approximations in regions with different distinguished limits and fixes their free constants by requiring agreement in an overlap region. The inner expansion resolves rapid variation; the outer expansion describes the surrounding region. Matching compares their common limiting behavior, rather than imposing distant boundary conditions directly on a local approximation.^[ref]

For the singular perturbation problem $ϵy^{\prime \prime }+y^{\prime }=0$, $y(0)=0$, $y(1)=1$, the leading outer solution is $y_0=1$. The inner variable $X=x/ϵ$ gives $Y_0^{\prime \prime }+Y_0^{\prime }=0$ and $Y_0(0)=0$, hence $Y_0=A(1-e^{-X})$. In the overlap region $ϵ\ll x\ll 1$, the inner variable tends to infinity while the outer coordinate tends to zero. The asymptotic matching condition isso $A=1$. One concrete path through the overlap region is $x=\sqrt{ϵ}$: then $x\to 0$ and $X={ϵ}^{-1/2}\to \infty$. Matching selects the free constant that the left boundary condition alone could not determine.

Why choose the right boundary condition for the outer solution? If instead one selects $y_0=0$ from the left, a proposed right boundary layer with $Z=(1-x)/ϵ$ satisfies $U_0^{\prime \prime }-U_0^{\prime }=0$. Its general solution is $A+B{e}^Z$. Matching to zero as $Z\to \infty$ forces $A=B=0$, contradicting $U_0(0)=1$. The direction in which the exponential function decays determines which endpoint can carry the boundary layer; endpoint selection is part of the calculation.

Split the integral at $ϵ\ll \delta \ll 1$. Replacing the numerator by one below the cutoff gives ${ϵ}^{-1}\arctan (\delta /ϵ)$, with an error bounded by $\frac{1}{2}\log (1+{\delta }^2/{ϵ}^2)$. Above the cutoff, replacing the denominator by $x^2$ has relative error $O({ϵ}^2/{\delta }^2)$, and integration by parts gives $1/\delta +\log \delta +O(1)$. The inner $-1/\delta$ cancels the outer $1/\delta$, and all errors are $o(1/ϵ)$, proving the leading constant independently of the intermediate cutoff.

For $ϵy^{\prime \prime }+x^{\alpha }{y}^{\prime }+y=0$ near zero, dominant balance gives a diffusion-drift layer of thickness ${ϵ}^{1/(\alpha +1)}$ for $-1<\alpha <1$, with the reaction term negligible. At $\alpha =1$, all three terms survive on the scale $\sqrt{ϵ}$. For $\alpha >1$, diffusion and reaction first balance on wavelength $\sqrt{ϵ}$, while the drift damps the oscillation on the larger envelope scale ${ϵ}^{1/(\alpha +1)}$. Removing the first derivative gives the envelope factor $\exp [-x^{\alpha +1}/(2ϵ(\alpha +1))]$. For $\alpha \le -1$, no shrinking algebraic diffusion layer of this regular bounded type is obtained.

For a positive small parameter $ϵ$, an order-one mismatch between the reduced solution $y_0=f(x)$ and a prescribed right endpoint value $y(1)=b$ creates a right boundary layer. Balancing $ϵy^{\prime \prime }$ against the order-one reaction term gives width $\sqrt{ϵ}$. In the inner variable $X=(1-x)/\sqrt{ϵ}$, the leading equation is $Y_0^{\prime \prime }=Y_0-f(1)$. Its boundary and matching conditions are $Y_0(0)=b$ and $Y_0\to f(1)$ as $X\to \infty$, soThe leading additive composite expansion is $f(x)+(b-f(1))e^{-(1-x)/\sqrt{ϵ}}$, obtained by subtracting the common overlap value. For the nonlinear perturbation $ϵy^{\prime \prime }=y-x^2+ϵ(2+y^2)$ with zero endpoint values, the first two outer expansion terms are $x^2-ϵx^4$, and the leading right-layer term is $1-e^{-X}$.

For $ϵy^{\prime \prime }-(1+x)y^{\prime }+y=0$, $y(0)=1$, $y(1)=1+ϵ$, the exact outer solution is $1+x$ and the mismatch is at the right endpoint. In $X=(1-x)/ϵ$, matching gives $Y_0=2-e^{-2X}$ and $Y_1=-X+(1-X-X^2/2)e^{-2X}$. Differentiation at $X=0$ gives the displayed endpoint slope. The order-one term requires the first inner correction; differentiating only the leading boundary layer loses it.

A growing linear mode can make a formally small derivative nonlinearity order one at large distance. After the crossover scaling, the equation $X{H}^{\prime \prime }-H^{\prime }+(H^{\prime }{)}^2=0$ reduces to a logistic differential equation in $\log X$ for $H^{\prime }$. Matching $H\sim X^2/2$ at small $X$ gives $H^{\prime }=X/(1+X)$ and the displayed primitive with $H(0)=0$. It changes from quadratic growth to linear growth without a finite turning singularity.

When the early growing solution is $A{x}^2$ and the nonlinear coefficient depends on $\epsilon y^{\prime }$, its crossover occurs at the displayed scale. If $A$ is order one this is $x=O({\epsilon }^{-1})$; if the initial slope suppresses the leading mode and $A=O(\epsilon )$, it is delayed to $x=O({\epsilon }^{-2})$. The first nonzero perturbative seed must therefore be computed before constructing the global leading approximation; setting the small parameter to zero can erase the mode that ultimately dominates.

For $q>0$, matching $Y\to -q$, $Y^{\prime }\to 0$ fixes the first integral to $(Y^{\prime }{)}^2=(q-2Y)(Y+q{)}^2/3$. Hence endpoint values above $q/2$ cannot match this outer branch. The upper family is $Y=-q+(3q/2){\mathrm{sech}}^2[\sqrt{q}(\xi -{\xi }_0)/2]$. An endpoint value strictly between $-q$ and $q/2$ admits two translations, one initially approaching $-q$ and one initially moving towards the turning value $q/2$. Values below $-q$ give the unique lower family $Y=-q-(3q/2){\mathrm{csch}}^2[\sqrt{q}(\xi +{\xi }_0)/2]$ with ${\xi }_0>0$. Matching to zero instead gives $(Y^{\prime }{)}^2=-q{Y}^2-2Y^3/3$, which forbids a nonconstant real approach to zero. These are leading boundary layer classifications, not exact multiplicity theorems at degenerate endpoint data.

In $ϵy^{\prime \prime }+xy+y^2=0$, balancing all three terms where the outer branches $0$ and $-x$ meet gives the displayed dominant balance and the inner equation $Y^{\prime \prime }+XY+Y^2=0$. Matching to $-X$ and writing $Y=-X+u$ gives $u^{\prime \prime }-Xu+u^2=0$. Its linearized decaying correction is an Airy function; the growing correction is excluded. Matching to zero instead gives $Y^{\prime \prime }+XY=0$, whose nonzero Airy function combinations oscillate. Thus the acceptable matching branches differ qualitatively at the turning endpoint.

An asymptotic matching condition requires the inner expansion and outer expansion, re-expressed in a common overlap region, to agree to the retained order. Both function values and slopes may constrain the free constants. In the square-root reaction problem with nested boundary layers, the shared small plateau sets the first integral of a square-root reaction layer, while its large-field fourth-power tail fixes the neighboring broad layer.

For fixed $k>0$ and endpoint values $y(0)=0$, $y(1)=1$, the outer solution is $y\sim {\epsilon }^2{k}^2$. The left boundary layer has width $\epsilon$. At the right, a layer of width $\sqrt{\epsilon }$ carries the order-one rise, and a translated layer of width $\epsilon$ connects its inner edge to the small plateau. The dominant balance depends on both amplitude and derivative scale; using the same scale at both ends loses a necessary match.

After $y={\epsilon }^{2}Y$ and $x=\epsilon \xi$ or a translated version, the layer equation is $Y^{\prime \prime }-\sqrt{Y}+k=0$. Multiplying by $Y^{\prime }$ and matching $Y=k^2$, $Y^{\prime }=0$ fixes the energy constant to $k^3/3$ and gives the displayed factorization. The increasing branch below $k^2$ approaches it at positive infinity; the increasing branch above it leaves it at negative infinity and grows like ${\xi }^4/144$. This supplies both interfaces of the square-root reaction problem with nested boundary layers.

Multiplication by $x^2$ makes this example exactly integrable once: $ϵx^{2}y\prime +(x^2-2{ϵ}^3)y=C$. Its homogeneous solution is $\exp (-x/ϵ-2{ϵ}^2/x)$. A solution with $y(1)={ϵ}^3$ that remains uniformly bounded needs $C\sim {ϵ}^3$, hence $y(0)\sim -1/2$. The scales $x=O(ϵ)$ and $x=O({ϵ}^2)$ connect the small outer solution to the order-one endpoint value. Bounded homogeneous contributions remain invisible to algebraic matching at the right boundary.

A decaying algebraic tail in an inner asymptotic expansion can become a fractional-order outer correction after rescaling. For an inner tail proportional to $r^{-1/2}$ and outer variable $x=ϵr$, the outer correction is proportional to ${ϵ}^{1/2}{x}^{-1/2}$. If the next outer correction contains $ϵ\log x$, expressing it in the inner variable produces both $ϵ\log r$ and $ϵ\log ϵ$. Omitting the latter loses a term larger than a plain $O(ϵ)$ correction. Matching must include all these scales and constants, rather than only the leading algebraic powers.

In a singular first-order equation with coefficient $x+ϵ$, an outer correction proportional to $ϵ\log x$ cannot be evaluated at the endpoint. The inner scale $x=ϵX$ changes it to $ϵ\log ϵ+ϵ\log (1+X)$. Matching retains the logarithmic constant, which is larger than a plain order-epsilon correction. For $(x+ϵ)y^{\prime }=ϵy$, $y(1)=1$, the exact solution $[(x+ϵ)/(1+ϵ){]}^{ϵ}$ verifies the displayed endpoint asymptotic and its remainder.

The equation $ϵy^{\prime \prime }+x{y}^{\prime }+y=0$ integrates to $ϵy^{\prime }+xy=C$. Equal endpoint data $y(-1)=y(1)=A$ force $C=0$ and give the unique Gaussian solution above. Its central width is $\sqrt{ϵ}$, with amplitude $A{e}^{1/(2ϵ)}$. The exponentially large central mode is missed by an assumed bounded algebraic outer expansion $C/x$. Changes of order one relative to either endpoint value occur over distance $O(ϵ)$ from that endpoint.

For $ϵy^{\prime \prime }+s^2{y}^{\prime }+y=0$ on $s\ge 0$, removing the drift by $y=e^{-s^3/(6ϵ)}v$ givesThe transformed potential changes sign at $s_t\sim \sqrt{2}{ϵ}^{1/4}$. WKB approximation is oscillatory inside and exponential outside this turning region; an Airy turning-point connection formula applies over width ${ϵ}^{5/12}$. At the endpoint the wavelength is $\sqrt{ϵ}$, while the physical exponential prefactor changes on the envelope scale ${ϵ}^{1/3}$. Balancing only diffusion and drift and declaring a single ${ϵ}^{1/3}$ layer misses the zeroth-order term and the turning region. Matching constants may be singular at homogeneous boundary-value resonances.

For $ϵy^{\prime \prime }+a(x)y^{\prime }+b(x)y=0$ with smooth $a$ bounded positively away from zero, the first-order equation for the outer expansion is selected by the right boundary condition. The left mismatch is repaired on the scale $\xi =x/ϵ$. If $a(0)=a_0>0$, the leading inner transient is $B{e}^{-a_0\xi }$. Its next equation is $Y_1^{\prime \prime }+a_0{Y}_1^{\prime }=-a^{\prime }(0)\xi Y_0^{\prime }-b(0)Y_0$. Matching its constant and linear large-$\xi$ terms to the outer expansion permits an order-$ϵ$ composite approximation. A sign reversal of the drift places the transient at the opposite endpoint.

If a leading outer expansion diverges as $\log x$ and a derivative coefficient contains $x-\epsilon z$, balancing that nonlinear correction gives a layer larger than $O(\epsilon )$. Write $x=\epsilon LX$ with $L=\log (1/\epsilon )$ and subtract the large negative constant from $z$. Matching often introduces a $\log L$ correction and yields an endpoint slope of order $1/(\epsilon L)$. The balance must be derived from the actual equation; logarithmic divergence alone does not fix a universal layer width.

For $\epsilon x{w}^{\prime \prime }+w^{\prime }+2xw=0$ with left endpoint $x=\epsilon$, diffusion there is of size ${\epsilon }^2$. The distinguished coordinate is $z=(x-\epsilon )/{\epsilon }^2$, giving $(1+\epsilon z)W^{\prime \prime }+W^{\prime }+2{\epsilon }^3(1+\epsilon z)W=0$. Matching to one with zero left value gives $W=1-e^{-z}-\epsilon z(z+2)e^{-z}/2+O({\epsilon }^2)$. The layer width is different from the endpoint's distance from zero.

In $\epsilon y^{\prime \prime }+2x{y}^{\prime }-2xy=0$, the drift vanishes at the endpoint $x=0$. Balancing diffusion against drift gives the inner expansion coordinate $z=x/\sqrt{\epsilon }$. The leading layer equation is $Y^{\prime \prime }+2z{Y}^{\prime }=0$, with a bounded error function profile. A regular outer first-order calculation alone cannot impose both endpoint values.

If an outer expansion contains $-(\epsilon /2)\log x$ and the layer coordinate is $x=\sqrt{\epsilon }z$, its overlap contains $-(\epsilon /2)\log z-(\epsilon /4)\log \epsilon$. Matching forces an inner expansion term $-(\epsilon /4)\log \epsilon \mathrm{erf}z$ when that homogeneous profile tends to one in the overlap and vanishes at the boundary. The logarithm multiplies the matching profile, rather than an arbitrary constant.

An overlap region is a parameter-dependent range in which two neighboring asymptotic expansions have simultaneously ordered common limits. Between scales ${ϵ}^2$ and $ϵ$, it is ${ϵ}^2\ll x\ll ϵ$: the narrow coordinate tends to infinity while the wider coordinate tends to zero. Matching uses these limiting descriptions, not an assertion that either expansion is uniform at all values of the other's scaled variable.

An intermediate asymptotic region has a scale between two other scales and provides their connecting matched asymptotic expansion. It can be needed because a coefficient changes its dominant form before the narrowest region's highest-derivative balance is reached. For example, a coefficient $x+ϵ$ changes near $x=O(ϵ)$ even when the fast inner expansion has width $O({ϵ}^2)$. The intermediate solution must be matched at both ends rather than assigned an independent boundary value.

In ${\epsilon }^2{y}^{\prime \prime }+a(x)y^{\prime }+\cdots =f$, a simple zero $a(x_{\ast })=0$, $a^{\prime }(x_{\ast })\ne 0$, balances diffusion and advection on width $\epsilon$. With $a(x)\simeq 2(x-x_{\ast })$, the leading homogeneous inner equation is $Y^{\prime \prime }+2z{Y}^{\prime }=0$, whose bounded solutions are constants and the error function. Thus distinct order-one outer solutions can match across an interior transition. Reversing the diffusion sign makes the nonconstant homogeneous solution grow at both ends, changing the matching structure.

For $-{\epsilon }^2{y}^{\prime \prime }+2(x-x_{\ast })y^{\prime }+\cdots =f$ on an interval straddling $x_{\ast }$, the bounded leading interior solution cannot connect unequal constants. A common bulk value is instead selected by an inner solvability condition, and width-${\epsilon }^2$ endpoint layers enforce the two boundary values. At endpoints $x=0,2$ with $x_{\ast }=1$, their leading decaying factors are $e^{-2x/{\epsilon }^2}$ and $e^{-2(2-x)/{\epsilon }^2}$. The interior layer still corrects derivative mismatches at smaller amplitude.

A distinguished limit rescales variables so that two or more effects enter the leading asymptotic balance together. It identifies a region whose governing equation differs from the equations in adjacent scales.

An inner expansion resolves a region in which a stretched independent variable makes otherwise small effects enter the leading balance.

The singular perturbation equation $ϵ\sqrt{x}{u}^{\prime \prime }+u^{\prime }=0$ balances its derivatives on the width $x={ϵ}^{2}X$. The inner equation is $\sqrt{X}{U}^{\prime \prime }+U^{\prime }=0$, so $U^{\prime }=C{e}^{-2\sqrt{X}}$. Its solution with $U(0)=0$ and $U(\infty )=1$ is the displayed $H$, since ${\int }_0^{\infty }{e}^{-2\sqrt{q}}dq=1/2$. Near zero $H=2X-8X^{3/2}/3+O(X^2)$, so the first derivative is finite while the second is singular. Treat the differential equation on the open interval and impose the endpoint value by continuous extension.

An inner variable is a stretched coordinate such as $\xi =(x-x_0)/\delta (ϵ)$ that remains order one inside a thin region of width $\delta (ϵ)$. Its scale is chosen by a dominant balance, rather than by the size of a coefficient alone.

For the singular perturbation problem $ϵy^{\prime \prime }+y^{\prime }=0$ with $y(0)=0$, $y(1)=1$, set $X=x/ϵ$ and $Y_{ϵ}(X)=y_{ϵ}(ϵX)$. The chain rule givesThus both derivative terms survive in the inner differential equation $Y_{ϵ}^{\prime \prime }+Y_{ϵ}^{\prime }=0$. The exact domain is $0\le X\le 1/ϵ$, withFor bounded $X$, $Y_{ϵ}\to Y_0$ as $ϵ\to 0^{+}$: the inner expansion has a nonconstant limiting profile even though the outer solution is constant. The finite-$ϵ$ profiles nearly collapse onto $Y_0$ after rescaling, with their remaining difference coming from the normalization at the distant right endpoint.

An outer expansion describes the solution away from a thin or distant region resolved by an inner expansion. It is matched to that inner expansion in their common overlap region. Its range of validity must be specified: an approximation at each fixed positive $x$ need not have uniform convergence down to an endpoint.

For the singular perturbation problem $ϵy^{\prime \prime }+y^{\prime }=0$, $y(0)=0$, $y(1)=1$, try the asymptotic expansion $y\sim y_0+ϵy_1+{ϵ}^2{y}_2+\cdots$ at fixed $x>0$. Substitution gives $y_0^{\prime }=0$ and $y_n^{\prime }=-y_{n-1}^{\prime \prime }$ for $n\ge 1$. Expanding the right boundary condition gives $y_0(1)=1$ and $y_n(1)=0$ for $n\ge 1$. Therefore $y_0=1$ and every higher coefficient vanishes. The missing boundary layer is exponentially small at each fixed positive $x$, invisible to all these powers of $ϵ$.

The exact solution supplies a quantitative limit. For every fixed $a\in (0,1)$,This is uniform convergence on $[a,1]$. On $[0,1]$ the same supremum norm of the error is exactly one. The inner expansion supplies the endpoint information that this outer expansion cannot encode.

An additive composite expansion combines an inner expansion and an outer expansion by subtracting their common behavior in the overlap region:The subtraction removes the part counted twice. Near the boundary layer, the outer expansion cancels the overlap contribution; away from the boundary layer, the inner expansion cancels it. An error estimate is needed to establish accuracy across the whole domain.^[ref]

For the singular perturbation problem $ϵy^{\prime \prime }+y^{\prime }=0$, $y(0)=0$, $y(1)=1$, the outer solution is one, the leading inner expansion is $1-e^{-x/ϵ}$, and their overlap is one. Thus $y_{\mathrm{comp}}=1-e^{-x/ϵ}$. With $q=e^{-1/ϵ}$, comparison with the exact normalized solution givesThe supremum norm error is therefore exponentially small on the entire interval, even though the outer solution alone has error one. This is an error bound for function values; errors in derivatives have additional powers of ${ϵ}^{-1}$.

The composite satisfies the differential equation exactly but has right boundary condition $y_{\mathrm{comp}}(1)=1-q$. Dividing it by $1-q$ restores that boundary condition and recovers the exact solution. This example makes both the strength and the limit of matching explicit: it gives a global approximation, while finite-$ϵ$ normalization can still carry information beyond every algebraic order.

When inner and outer approximations share a nonzero multiplicative overlap, a multiplicative composite divides their product by that common part. It is especially convenient when different regions contribute separate exponential or boundary-layer factors.

For $ϵy^{\prime \prime }+a(x)y^{\prime }+b(x)y=0$ with $a>0$, the reduced first-order outer problem takes its boundary condition at the right endpoint. A rapidly decaying exponential in $x/ϵ$ supplies the otherwise unsatisfied left-end condition, so the layer lies at the outflow end for the fast mode associated with this sign convention.

For $ϵy^{\prime \prime }-y^{\prime }+e^y=0$ with $y(0)=A<0$ and $y(1)=0$, the outer solution is $-\log (e^{-A}-x)$. The right-end boundary layer has width $ϵ$ and leading correction $\log (e^{-A}-1)e^{(x-1)/ϵ}$. This follows by setting $X=(x-1)/ϵ$, solving $Y^{\prime \prime }-Y^{\prime }=0$, and matching its constant part. The leading correction changes sign at $A=-\log 2$.

A switchback term such as $ϵ\log ϵ$ is forced into a matched expansion when logarithms of a stretched coordinate split into parameter-dependent and local pieces in the overlap region.

If a stretched variable is $x=ϵr$ and an outer expansion contains ${ϵ}^2\log x$, its expression in the fixed-$r$ region is ${ϵ}^2\log r+{ϵ}^2\log ϵ$. The parameter logarithm must therefore be present in the inner expansion, often multiplying a homogeneous solution selected by a boundary condition. An ansatz containing only integer powers with parameter-independent coefficients misses this switchback term. The same reasoning applies to other powers of $ϵ$ and other stretched-coordinate scalings.

The method of multiple scales treats fast and slow variables such as $t$ and $T=ϵt$ as independent, then removes resonant forcing to obtain slow amplitude and phase equations.

For $x^{\prime \prime }+3ϵy{x}^{\prime }+x=2(y^{\prime }{)}^2$ and $y^{\prime }=ϵ(1+x-y)$ with leading oscillation $x=A(\tau )\sin t$, $\tau =ϵt$, averaging the auxiliary equation gives $Y^{\prime }=1-Y$. Cancellation of resonant forcing in the oscillator gives $2A^{\prime }+3YA=0$. With $Y(0)=3$ and $A(0)=1$, $Y=1+2e^{-\tau }$ and the displayed amplitude follows. Bounded oscillatory corrections repair the instantaneous auxiliary equation; secular terms in a fixed-time expansion show why the slow scale is needed.

For $y^{\prime \prime }+ϵ(y^{\prime }{)}^n+y=0$, the method of multiple scales gives no first-order phase drift. If $n$ is odd, the amplitude obeys the displayed damping law with $C_n=2^{-(n+1)}\left(\genfrac{}{}{0ex}{}{n+1}{(n+1)/2}\right)$. If $n$ is even, the average is zero and the amplitude is unchanged on the first slow time $T=ϵt$. The odd case dissipates mechanical energy, whereas the even case is reversible and its force is not damping of both velocity signs.

Two fundamental harmonics of frequency one multiply into only a constant and second harmonics. The first-order solvability condition in the method of multiple scales therefore has no amplitude-changing resonant coupling. Their first-order particular corrections can mix back into the fundamental at second order, selecting the possible scale $t=O({ϵ}^{-2})$. Detuning may suppress the resulting resonance, so identifying this scale alone does not prove instability for every initial state.

For $u^{\prime \prime }+u=ϵu^2$, a leading oscillation $U_0=A(T)e^{it}+\overline{A}(T)e^{-it}$ has slow time $T={ϵ}^{2}t$. The first-order forcing contains only zero and double frequencies. At second order the fundamental forcing is $(10/3)|A{|}^{2}A-2i{A}_T$; eliminating resonance gives the displayed equation. Its modulus is constant and its phase drifts. A leading real amplitude $R=2|A|$ therefore has frequency $1-5R^2{ϵ}^2/12+O({ϵ}^3)$. This replaces a secular phase term by a uniformly modulated oscillation.

For $x\prime \prime +ϵf(x)x\prime +x=0$, write $x_0=A(T)\cos (t+\theta )$ and $T=ϵt$. Removing secular terms gives $A_T=-A⟨f(A\cos \psi ){\sin }^2\psi ⟩$ and ${\theta }_T=0$, where the brackets mean a period average. This is the first-order solvability condition in the method of multiple scales, with a slowly evolving amplitude and no first-order frequency correction.

For $x^{\prime \prime }+\epsilon |x|x^{\prime }+x=0$ and unit initial amplitude, averaging uses $⟨|\cos \psi |{\sin }^2\psi ⟩=2/(3\pi )$. Thus $A_T=-2A^2/(3\pi )$ and the displayed algebraic envelope follows. The leading phase has no first-order drift. The dissipation is amplitude dependent, so an exponential envelope with a constant decay rate would give the wrong long-time scale.

For $x^{\prime \prime }+\epsilon (x^2-1)x^{\prime }+x=0$, the averaged amplitude for position-dependent damping has the displayed logistic-type equation. With $A(0)=1$, its positive solution is $A(T)=2/\sqrt{1+3e^{-T}}$. Small amplitudes grow and large ones decay toward the stable value two; the leading oscillation is $A(\epsilon t)\cos t$.

For $x^{\prime \prime }+ϵc(1-x^2)x^{\prime }+x=k$, use $T=ϵt$ and $x=k-r(T)\cos (t+\varphi (T))$. Removal of secular terms gives ${\varphi }_T=0$ and the displayed amplitude equation. With $R=r^2$, $R_T=c(k^2-1)R+(c/4)R^2$, which is explicitly integrable. If $c$ is replaced by a slowly varying coefficient $C(T)$, the same solution uses the accumulated time ${\int }_0^{T}C(s)ds$. The method of multiple scales is uniform on bounded slow-time intervals only while its amplitude and correction estimates remain bounded.