Indicial Equation and Exponents

At an ordinary point a linear second-order ODE is tame: two power series, two analytic solutions, no drama. Step onto a singular point — where the coefficients blow up — and that guarantee evaporates. The method of Frobenius rescues the series idea at the mildest singularities, the regular singular points, by conceding one thing: the solution need not start with a whole power. It may lead with $x^r$ for some exponent $r$ that is not an integer, not even real. The indicial equation is the small algebraic gatekeeper that decides which exponents are allowed. Substitute the ansatz $y=x^r\sum a_n x^n$ , collect the very lowest power of $x$ , and its coefficient must vanish — that single condition is a quadratic in $r$ , and its two roots are the exponents at the singularity. Everything downstream hangs on those two numbers. Their difference $r_1-r_2$ decides whether the second solution is another clean series or is forced to carry a $\ln x$ term — a logarithm that appears not by choice but because reduction of order, applied to a single Frobenius solution, integrates a $1/x$ into existence. The Wronskian and existence theory promise exactly two independent solutions; the indicial analysis tells you their shape near the singular point. And this is no formal curiosity: Bessel's equation hands you exponents $\pm\nu$ , Legendre's hands you $0$ and $1$ , the hydrogen radial equation hands you $\ell+1$ and $-\ell$ , and physical normalizability quietly selects one of each pair. Master the indicial equation and you hold the local key to nearly every special function in mathematical physics.

Key takeaways

A regular singular point admits Frobenius solutions $y=x^{r}\sum_{n\ge0}a_n x^n$ whose leading exponent $r$ is fixed by the indicial equation, the local analogue of a characteristic equation.
The indicial equation is the quadratic $r(r-1)+p_0 r+q_0=0$ , where $p_0=\lim_{x\to0}xp(x)$ and $q_0=\lim_{x\to0}x^2q(x)$ in the standard form $y''+py'+qy=0$ .
Its two roots $r_1\ge r_2$ — the exponents — satisfy Vieta's relations $r_1+r_2=1-p_0$ and $r_1r_2=q_0$ , and seed the recurrence that builds every higher coefficient.
When $r_1-r_2\notin\mathbb{Z}$ both roots give independent series; a repeated root always forces a $\ln x$ second solution, and an integer difference may.
Worked to numbers: Bessel order $\tfrac12$ has exponents $\pm\tfrac12$ (difference $1$ , yet no log — the solutions are $\sin x/\sqrt{x}$ and $\cos x/\sqrt{x}$ ), while the Euler equation $x^2y''+3xy'+y=0$ has the repeated root $-1$ and the exact second solution $x^{-1}\ln x$ .
Key idea: substitute $y=\sum a_n x^{n+r}$ into the ODE and demand the coefficient of the lowest power vanish with $a_0\ne0$ ; this gives the indicial quadratic $r(r-1)+p_0r+q_0=0$ , whose roots are the local exponents.

The core insight

Substitute $y=\sum a_n x^{n+r}$ into the ODE and demand that the coefficient of the lowest power vanish with $a_0\ne0$ : this gives the indicial quadratic $r(r-1)+p_0r+q_0=0$ , whose two roots are the exponents at the regular singular point — the local powers with which every solution must begin.

Where this lives today

The indicial equation is not a museum piece — it is the first line of almost every serious calculation with a special function. In atomic and nuclear physics, solving the radial Schrödinger or Dirac equation begins by extracting the exponent at the origin: for a Coulomb potential it is $\ell+1$ (Schrödinger) or the relativistic $\gamma=\sqrt{(\kappa)^2-(Z\alpha)^2}$ (Dirac), and that non-integer $\gamma$ is directly responsible for the fine-structure shift of hydrogenic energy levels. Computational packages that evaluate hypergeometric and Heun functions — the workhorses of black-hole perturbation theory, where the radial Teukolsky equation has regular singular points at the horizon and infinity — build their connection coefficients from the exponents supplied by the indicial equation at each singular point. In condensed matter, the local exponents of the Schrödinger equation near an impurity or a band edge set the threshold behaviour of scattering and the shape of van Hove singularities. And in the theory of automatic ODE solvers and computer algebra, Fuchs's and Frobenius's classification is exactly what lets a symbolic engine decide, before integrating, whether a point is ordinary, regular-singular, or irregular — routing the problem to a Taylor series, a Frobenius series, or an asymptotic expansion accordingly. Every time a code prints "solution $\sim x^{\ell+1}$ near the origin," an indicial equation was quietly solved.

The physical picture

Near a regular singular point a solution behaves like $x^r$ to leading order, exactly as $e^{\lambda t}$ leads in a constant-coefficient equation. The indicial equation is the local characteristic equation: a quadratic whose roots are the admissible leading exponents. The Frobenius factor $x^r$ absorbs the singular behaviour, and the analytic series $\sum a_n x^n$ mops up the rest. The first figure draws that quadratic for a concrete case — Bessel's equation of order two, where the indicial polynomial is simply $F(r)=r^2-4$ and its roots sit at $r=\pm2$ .

Figure

The indicial equation is the characteristic equation of a singular point. For Bessel order ν = 2 it reads F(r) = r(r−1) + p₀r + q₀ = r² − 4, a parabola whose two zero crossings are the exponents r₁ = +2 and r₂ = −2. The roots are where the curve meets the axis; their sum 1 − p₀ = 0 and product q₀ = −4 are read straight off Vieta.

The two roots do more than exist — their difference dictates the character of the whole solution space, and that character has direct physical teeth. Consider the reduced radial Schrödinger equation near the origin, $u''-\ell(\ell+1)x^{-2}u\approx0$ : its exponents are $\ell+1$ and $-\ell$ . Both are honest solutions of the differential equation, but only one keeps the wavefunction finite and normalizable at the origin. The indicial equation lays both options on the table; physics picks the regular one. The second figure shows the two competing local behaviours for $\ell=1$ — the exponent $+2$ curve that vanishes smoothly at the origin and the exponent $-1$ curve that diverges and is thrown away.

Figure

Exponent selection near a singular point, hydrogen reduced radial equation with ℓ = 1. The indicial roots are ℓ + 1 = 2 and −ℓ = −1. The regular branch u ∼ x² vanishes at the origin and is normalizable; the irregular branch u ∼ x⁻¹ diverges and is discarded. Both solve the ODE — physics keeps only the exponent that gives finite probability.

The core equations

The starting point is the normalized second-order equation. A point $x=0$ is a regular singular point when the coefficients are singular there but not too badly: $p$ may have at most a simple pole and $q$ at most a double pole, which is the same as saying $xp$ and $x^2q$ stay analytic.

y''+p(x)\,y'+q(x)\,y=0,\qquad xp(x)=\sum_{k\ge0}p_k x^{k},\qquad x^2q(x)=\sum_{k\ge0}q_k x^{k} \qquad (1)

where: $p,q$ — coefficient functions, singular at $x=0$ ; $p_k,q_k$ — Taylor coefficients of the analytic functions $xp$ and $x^2q$ ; $p_0,q_0$ — the leading values $p_0=\lim_{x\to0}xp(x)$ , $q_0=\lim_{x\to0}x^2q(x)$ . Reading: only $p_0$ and $q_0$ enter the indicial equation — never $p(0)$ or $q(0)$ , which generally diverge. The Frobenius ansatz replaces the ordinary power series with a series multiplied by a floating power $x^r$ . The normalization $a_0\ne0$ is essential: it is what pins $x^r$ as the genuine leading term.

y=x^{r}\sum_{n=0}^{\infty}a_n x^{n},\qquad a_0\ne0 \qquad (2)

where: $r$ — the exponent, to be fixed by the indicial equation; $a_n$ — series coefficients; $a_0\ne0$ — normalization forcing $x^r$ to lead. Reading: if $r$ is not an integer the solution is genuinely non-analytic at $x=0$ , which is exactly why an ordinary Taylor series failed. Substituting the ansatz and demanding the lowest surviving power of $x$ vanish yields the central result. It is a quadratic in $r$ built from just two numbers.

r(r-1)+p_0\,r+q_0=0,\qquad F(r)\equiv r^2+(p_0-1)r+q_0 \qquad (3)

where: $r$ — exponent; the two roots are the indicial exponents; $F(r)$ — the indicial polynomial, whose value at $n+r$ later gates the recurrence. Reading: this is the local characteristic equation — compare $F(r)=0$ with $m\lambda^2+c\lambda+k=0$ for a constant-coefficient oscillator. Solving the quadratic gives the exponents explicitly, and Vieta's formulas let you read their sum and product off the coefficients without solving at all.

r_{1,2}=\tfrac12\Big[(1-p_0)\pm\sqrt{(1-p_0)^2-4q_0}\Big],\qquad r_1+r_2=1-p_0,\qquad r_1 r_2=q_0 \qquad (4)

where: $r_1,r_2$ — exponents, conventionally $\operatorname{Re}r_1\ge\operatorname{Re}r_2$ . Reading: the discriminant $(1-p_0)^2-4q_0$ decides whether the exponents are real and distinct, complex, or coincident — the local version of over-, under-, and critically-damped. Beyond the leading power, every higher coefficient is fixed by a recurrence in which the indicial polynomial reappears as the prefactor of the newest unknown.

F(n+r)\,a_n=-\sum_{k=0}^{n-1}\big[\,p_{n-k}(k+r)+q_{n-k}\,\big]a_k \qquad (5)

where: $F(n+r)$ — indicial polynomial evaluated at the shifted exponent; $a_k$ — earlier coefficients on the right. Reading: as long as $F(n+r)\ne0$ you divide and march on; the whole theory of logarithms is the story of when $F(n+r)=0$ . When the recurrence breaks down — a repeated root, or an integer-spaced pair where the right-hand side refuses to vanish — the second solution acquires a logarithm.

y_2(x)=C\,y_1(x)\ln x+x^{r_2}\sum_{n=0}^{\infty}b_n x^{n},\qquad b_0\ne0 \qquad (6)

where: $C$ — coupling constant; $C\ne0$ is forced for a repeated root and possible for an integer difference; $b_n$ — second-solution coefficients. Reading: the log is never an ansatz plucked from air — it is the residue of reduction of order, derived below. Finally, summing the exponents over all regular singular points of a Fuchsian equation gives a global constraint that no solution can violate.

\sum_{j}\big(\rho_j+\sigma_j\big)=N-1 \qquad (7)

where: $\rho_j,\sigma_j$ — the two exponents at the $j$ -th regular singular point (the point at infinity counts); $N$ — number of regular singular points. Reading: for the hypergeometric equation ( $N=3$ ) this fixes the sum of all six exponents at $2$ , a check every table of exponents must pass.

Derivation ADeriving the indicial equation from the Frobenius ansatz

The goal is to turn the vague instruction "substitute a series and collect the lowest power" into the exact quadratic $r(r-1)+p_0r+q_0=0$ , making visible precisely which terms survive at leading order and why the normalization $a_0\ne0$ is what converts a coefficient condition into an equation for $r$ . We multiply into analytic form, insert the ansatz, and isolate the single power of $x$ that every term shares at the bottom.

Step 1. Multiply the standard form by $x^2$ to clear the poles: $x^2y''+x\,[xp(x)]\,y'+[x^2q(x)]\,y=0$ , where by definition $xp=\sum_k p_k x^k$ and $x^2q=\sum_k q_k x^k$ are analytic. Why: in this shape every coefficient is a convergent power series starting at $x^0$ , so products with the ansatz have a clean lowest power — the singular original form does not.

Step 2. Differentiate the ansatz term by term, $y'=\sum_n(n+r)a_nx^{n+r-1}$ and $y''=\sum_n(n+r)(n+r-1)a_nx^{n+r-2}$ , and insert. The three pieces become $\sum_n(n+r)(n+r-1)a_nx^{n+r}$ , $\big(\sum_kp_kx^k\big)\sum_n(n+r)a_nx^{n+r}$ , and $\big(\sum_kq_kx^k\big)\sum_na_nx^{n+r}$ . Why: after the $x^2$ multiply each term now begins at the common power $x^{r}$ , so they can be compared coefficient by coefficient.

Step 3. Extract the coefficient of the lowest power $x^{r}$ , which comes only from $n=0$ and $k=0$ : $\big[r(r-1)+p_0r+q_0\big]a_0=0$ . Every other contribution carries at least one extra factor of $x$ . Why: the lowest-order balance cannot borrow from higher coefficients, so it is a closed condition on $r$ alone once $a_0$ is fixed.

Step 4. Invoke the normalization. Since $a_0\ne0$ by assumption, the bracket must vanish: $r(r-1)+p_0r+q_0=0$ . Why: had we allowed $a_0=0$ the "leading" term would be a lie — the series would really start higher up — so $a_0\ne0$ is exactly what makes $x^r$ the true leading power and forces the quadratic. *For Bessel's equation, $p=1/x$ gives $p_0=1$ and $q=1-\nu^2/x^2$ gives $q_0=-\nu^2$ , so the indicial polynomial is $F(r)=r^2-\nu^2=0$ and the exponents are $r=\pm\nu$ — for instance $r=\pm2$ when $\nu=2$ and $r=\pm\tfrac12$ when $\nu=\tfrac12$ .*

Go deeper · Derivation B · 1 minWhy a repeated root forces a logarithm

Derivation BWhy a repeated root forces a logarithm

The goal is to show that when $r_1=r_2$ the second solution cannot be a pure Frobenius series — a $\ln x$ term is mathematically unavoidable — and to see it emerge from reduction of order rather than being assumed. We take the one Frobenius solution the repeated root supplies, feed it into the reduction-of-order integral, and track the leading power through Abel's formula until the logarithm falls out on its own.

Step 1. With one solution $y_1=x^{r}\sum a_nx^n$ in hand, reduction of order gives a second as $y_2=y_1\int^x \dfrac{e^{-\int p\,\mathrm{d}t}}{y_1^2}\,\mathrm{d}s$ , the Abel construction. Why: for a second-order linear ODE this integral always produces a genuinely independent solution, so any structure it forces — including a log — is forced on the true $y_2$ .

Step 2. Evaluate Abel's factor. Since $p=p_0/x+\cdots$ , we get $e^{-\int p\,\mathrm{d}x}=x^{-p_0}\times(\text{analytic},=1+\cdots)$ , and for the repeated root $r_1+r_2=1-p_0$ with $r_1=r_2=r$ gives $p_0=1-2r$ . Why: only the leading $p_0/x$ controls the singular power of the integrand; the analytic remainder cannot change which power sits at the bottom.

Step 3. Assemble the integrand: $\dfrac{x^{-p_0}}{y_1^2}\sim\dfrac{x^{-(1-2r)}}{x^{2r}}=\dfrac{1}{x}$ , times an analytic series $1+c_1x+\cdots$ . Why: the exponents cancel to give exactly $x^{-1}$ — the one power whose integral is not another power but a logarithm; a repeated root is precisely the condition that produces this cancellation.

Step 4. Integrate: $\int\!\big(\tfrac1x+c_1+c_2x+\cdots\big)\mathrm{d}x=\ln x+(\text{power series})$ , so $y_2=y_1\ln x+x^{r}\sum_n b_nx^n$ . Why: the $x^{-1}$ term integrates to $\ln x$ and nothing can remove it, so the second solution of a repeated-root singularity always carries a logarithm with coupling $C=1$ in this normalization. *The Euler equation $x^2y''+3xy'+y=0$ has the repeated root $r=-1$ and second solution $x^{-1}\ln x$ , as confirmed by direct substitution; the Bessel case $\nu=0$ (repeated root $0$ ) likewise reproduces the logarithm in $Y_0$ .*

Go deeper · Derivation C · 1 minThe integer-difference case and when the logarithm vanishes

Derivation CThe integer-difference case and when the logarithm vanishes

The goal is the subtler middle ground: when $r_1-r_2=N$ is a positive integer, the smaller root's recurrence hits a wall at step $N$ , and whether a logarithm appears depends on a single computable sum. We locate the resonance, show why the coefficient of $a_N$ dies, and reduce the log-or-no-log question to the vanishing of the accumulated right-hand side.

Step 1. Build the series for the smaller root $r_2$ and reach step $n=N$ : the recurrence reads $F(N+r_2)\,a_N=-\sum_{k=0}^{N-1}[\,p_{N-k}(k+r_2)+q_{N-k}\,]a_k$ . Why: this is the first step where the earlier coefficients have all been fixed, so the outcome depends only on their accumulated value against the prefactor.

Step 2. Show the prefactor vanishes. Because $N+r_2=r_1$ is the other indicial root and $F$ vanishes at both roots, $F(N+r_2)=F(r_1)=0$ . Why: the recurrence tries to solve $0\cdot a_N=(\text{right side})$ , which is either impossible or leaves $a_N$ free — the same resonance that a repeated root shows, now displaced to step $N$ .

Step 3. Read off the two outcomes. If the right-hand sum is zero, then $0=0$ holds, $a_N$ is arbitrary, and a pure Frobenius series exists for $r_2$ with $C=0$ . If the sum is nonzero, no finite $a_N$ works, consistency fails, and a term $C\,y_1\ln x$ with $C\ne0$ must be added. Why: the logarithm is the minimal repair that restores solvability when the resonant equation is otherwise contradictory.

Step 4. Reduce it to a test: $C\propto\sum_{k=0}^{N-1}[\,p_{N-k}(k+r_2)+q_{N-k}\,]a_k$ . Compute this one sum and its (non)vanishing decides everything. Why: it converts a qualitative "may or may not have a log" into a definite arithmetic check on already-known coefficients. *For Bessel order $\tfrac12$ , $r_1-r_2=1$ yet the resonant sum vanishes — the recurrence couples $n$ to $n-2$ , so step $1$ has an empty right side — and no logarithm appears; indeed $J_{\pm1/2}\propto x^{-1/2}\sin x,\ x^{-1/2}\cos x$ are elementary. By contrast, integer-order Bessel $\nu=N$ has a nonzero resonant sum and $Y_N$ carries a logarithm.*

Limiting cases

With roots $r_1\ge r_2$ , the entire structure of the solution space is governed by the difference $r_1-r_2$ and by whether the singularity is regular at all.

Case	Result	Meaning
Distinct, non-integer difference	$r_1-r_2\notin\mathbb{Z}$	Two independent Frobenius series $y_i=x^{r_i}\sum a_n^{(i)}x^n$ ; no logarithm needed — the generic and simplest case.
Repeated root	$r_1=r_2=r$	One series $y_1=x^{r}\sum a_nx^n$ ; the second always carries a log, $y_2=y_1\ln x+x^{r}\sum b_nx^n$ .
Difference a positive integer	$r_1-r_2=N\in\mathbb{Z}^{+}$	$y_1$ from $r_1$ always works; $y_2$ from $r_2$ needs $C\,y_1\ln x$ only if the resonant sum at $n=N$ is nonzero.
Complex conjugate exponents	$(1-p_0)^2<4q_0$	$r=\alpha\pm i\beta$ ; solutions $x^\alpha\cos(\beta\ln x)$ and $x^\alpha\sin(\beta\ln x)$ oscillate infinitely as $x\to0$ .
Irregular singular point	$x^2q(x)\to\infty$	The indicial method fails outright; solutions may carry essential singularities — asymptotic or WKB analysis is required.

Worked examples

1 · Bessel order one-half — an integer difference with no logarithm (fully worked)

Find the exponents and the two series solutions of $y''+\dfrac{1}{x}y'+\Big(1-\dfrac{1}{4x^2}\Big)y=0$ , and decide whether a logarithm appears.

Step 1. Read the indicial data. Here $xp=1$ so $p_0=1$ , and $x^2q=x^2-\tfrac14$ so $q_0=-\tfrac14$ . The indicial polynomial is $F(r)=r(r-1)+r-\tfrac14=r^2-\tfrac14$ , giving exponents $r_1=+\tfrac12$ and $r_2=-\tfrac12$ . Their difference is $r_1-r_2=1$ , a positive integer — so a logarithm is possible and must be checked, not assumed.

Step 2. Build the recurrence. Only $q_2=1$ couples different orders, so equation (5) collapses to $F(n+r)\,a_n=-a_{n-2}$ , i.e. $\big[(n+r)^2-\tfrac14\big]a_n=-a_{n-2}$ . This links $a_n$ only to $a_{n-2}$ , so odd and even coefficients never mix.

Step 3. Take the larger root $r=+\tfrac12$ . Then $F(n+\tfrac12)=(n+\tfrac12)^2-\tfrac14=n(n+1)$ , so $a_n=-a_{n-2}/[n(n+1)]$ . With $a_0=1$ : $a_2=-\tfrac{1}{6}$ , $a_4=+\tfrac{1}{120}$ , giving $y_1=x^{1/2}\big(1-\tfrac{x^2}{6}+\tfrac{x^4}{120}-\cdots\big)=x^{-1/2}\sin x$ . The bracketed series is exactly $\sin x/x$ .

Step 4. Take the smaller root $r=-\tfrac12$ , the resonant one. Now $F(n-\tfrac12)=n(n-1)$ , which vanishes at $n=1$ . But the recurrence at $n=1$ reads $0\cdot a_1=-a_{-1}=0$ : the right side is empty, so the resonance is harmless and $a_1$ is free (take $a_1=0$ ). Continuing, $a_2=-\tfrac12$ , $a_4=+\tfrac{1}{24}$ , so $y_2=x^{-1/2}\big(1-\tfrac{x^2}{2}+\tfrac{x^4}{24}-\cdots\big)=x^{-1/2}\cos x$ .

Result. The exponents are $\pm\tfrac12$ ; despite the integer difference there is no logarithm, and the two independent solutions are $y_1=\sin x/\sqrt{x}$ and $y_2=\cos x/\sqrt{x}$ — the elementary form of $J_{1/2}$ and $J_{-1/2}$ . This is the textbook warning made concrete: an integer difference is necessary for a log, never sufficient.

2 · The Euler equation — a repeated root and an exact logarithm (fully worked)

Solve $x^2y''+3xy'+y=0$ near its regular singular point $x=0$ .

Step 1. An Euler (equidimensional) equation has $xp=3$ and $x^2q=1$ constant, so $p_0=3$ , $q_0=1$ , and every higher $p_k,q_k=0$ . The indicial polynomial is therefore the entire story — there is no series to build.

Step 2. Form the indicial equation: $F(r)=r(r-1)+3r+1=r^2+2r+1=(r+1)^2=0$ . The single repeated root is $r=-1$ . The discriminant $(1-p_0)^2-4q_0=4-4=0$ confirms the coincidence.

Step 3. The first solution is $y_1=x^{-1}$ (the series truncates at one term because all $p_k,q_k$ beyond leading order vanish). A repeated root forbids a second pure power, so by the reduction-of-order result the second solution is $y_2=x^{-1}\ln x$ .

Step 4. Verify directly. With $y_2=x^{-1}\ln x$ : $y_2'=-x^{-2}\ln x+x^{-2}$ and $y_2''=2x^{-3}\ln x-3x^{-3}$ . Then $x^2y_2''+3xy_2'+y_2=(2-3+1)x^{-1}\ln x+(-3+3)x^{-1}=0$ .

Result. The general solution is $y=c_1\,x^{-1}+c_2\,x^{-1}\ln x$ — the logarithm here is exact, not the leading term of an infinite series, which makes the Euler equation the cleanest possible demonstration of why a repeated root forces $\ln x$ .

Physics of the portrait

The indicial equation is the crystallized legacy of a nineteenth-century programme to tame differential equations by studying their singular points — a programme that turned a case-by-case art into the systematic theory behind every special function a physicist uses.

Origins · 1 minGeorg Frobenius and how this was discovered

Georg Frobenius

1849–1917 · series solutions at regular singular points

Ferdinand Georg Frobenius was a Berlin mathematician of formidable range — group theory, matrix algebra, and number theory all bear his name — but for the working physicist his 1873 paper Über die Integration der linearen Differentialgleichungen durch Reihen is the enduring gift. Building on Lazarus Fuchs's classification of singular points, Frobenius asked a sharp question: given a regular singular point, can one always write down solutions as generalized power series, and what happens when the naive recipe breaks? His answer is the method now universally taught in his name. He showed that the leading exponent is governed by a quadratic — the indicial equation — and, crucially, he resolved the awkward cases where the two exponents coincide or differ by an integer, deriving the logarithmic second solution rather than merely noting its necessity. What makes Frobenius's treatment definitive is its completeness: where earlier writers handled special equations one at a time, he gave a single algorithm valid for the entire class of Fuchsian equations, reducing the local behaviour of Bessel, Legendre, hypergeometric, and Laguerre functions to one recurrence and one quadratic. A rigorous stylist who distrusted geometric hand-waving, Frobenius insisted on convergence proofs that later generations could build on. The mathematics of the hydrogen atom, the vibrating drum, and the multipole expansion all pass through the small gate he built.

History

1785

— Legendre introduces the polynomials Pₗ that solve his equation on [−1, 1]

1824

— Bessel systematically studies the functions Jₙ arising in planetary motion

1857

— Riemann encodes a Fuchsian equation entirely by its exponents in the P-symbol

1866

— Fuchs defines regular singular points and the exponents attached to them

1873

— Frobenius gives the complete series method, resolving the logarithmic cases

Going further

Texts. Arfken & Weber, Mathematical Methods for Physicists, 7th ed., ch. on Series Solutions and the Frobenius method · Hassani, Mathematical Physics, ch. on second-order linear ODEs and Fuchsian equations · Stone & Goldbart, Mathematics for Physics, §16 (Frobenius method, indicial equation) · Byron & Fuller, Mathematics of Classical and Quantum Physics, ch. 2 · Riley, Hobson & Bence, Mathematical Methods for Physics and Engineering, §16.3 · Bender & Orszag, Advanced Mathematical Methods for Scientists and Engineers, §3.3. The indicial quadratic $r(r-1)+p_0r+q_0=0$ , the root and Vieta formulas, and the recurrence (5) follow from the Frobenius ansatz. Bessel order $\tfrac12$ yields $y_1=\sin x/\sqrt{x}$ , $y_2=\cos x/\sqrt{x}$ with no logarithm; the Euler equation $x^2y''+3xy'+y=0$ has the exact repeated-root pair $x^{-1},\,x^{-1}\ln x$ , both confirmed by direct substitution. The hydrogen exponents $\ell+1,-\ell$ (reduced radial) and the Bessel exponents $\pm\nu$ are standard results. For Bessel, $p_0=1$ and $q_0=-\nu^2$ give $F(r)=0$ with roots $r=\pm\nu$ , satisfying $r_1+r_2=1-p_0$ and $r_1r_2=q_0$ ; the resonance condition $F(N+r_2)=0$ governs the integer-difference cases treated above.

Prerequisites. The Method of Frobenius · Ordinary and Singular Points · Power Series Solutions · Linear Second-Order ODEs · Reduction of Order. Leads to. Sturm–Liouville Theory · Two-Point Boundary-Value Problems · The Wronskian and Linear Independence.

Ask the tutor: Hand me any second-order ODE and a point, and I will test whether it is a regular singular point, write down its indicial equation, and tell you exactly which solutions carry a logarithm.

Why does the ansatz start with x^r instead of an ordinary power series?

Because at a singular point an ordinary Taylor series simply cannot represent the solution. The coefficients $p$ and $q$ blow up there, and a function forced to match them may itself behave like $x^{1/2}$ or $x^{-\nu}$ or $\ln x$ near the origin — none of which is a whole-power series. The floating exponent $r$ gives the ansatz exactly enough freedom to capture that leading singular behaviour, after which the ordinary analytic series $\sum a_nx^n$ can handle the smooth remainder. At an ordinary point the mechanism still works but returns $r=0$ and $r=1$ , recovering the two Taylor solutions — so Frobenius contains the ordinary-point theory as a special case.

Which exponent do I use to build the first solution?

Always start with the larger root $r_1$ (larger real part). Its recurrence never hits the resonance $F(n+r_1)=0$ for $n\ge1$ , because $n+r_1$ can never equal the smaller root $r_2$ when $n>0$ — so the series for $r_1$ is always clean and always exists. The smaller root $r_2$ is where trouble may live: if $r_1-r_2$ is a positive integer, its recurrence can stall at step $N=r_1-r_2$ . Build $y_1$ from $r_1$ first, unconditionally, then attempt $y_2$ from $r_2$ and check the resonance.

How is the indicial equation related to the characteristic equation of a constant-coefficient ODE?

They are the same idea in two settings. For $ay''+by'+cy=0$ you substitute $y=e^{\lambda t}$ and get the algebraic characteristic equation $a\lambda^2+b\lambda+c=0$ ; its roots are the growth rates. For a regular singular point you substitute $y=x^r(1+\cdots)$ and get the indicial equation $r^2+(p_0-1)r+q_0=0$ ; its roots are the local powers. The parallels run deep: distinct roots give two independent modes, a repeated root spawns an extra factor ( $te^{\lambda t}$ there, $\ln x$ here), and complex roots give oscillation ( $e^{\alpha t}\cos\beta t$ there, $x^\alpha\cos(\beta\ln x)$ here). The indicial equation is the characteristic equation localized at a singularity.

Can the exponents be complex, and what would that mean physically?

Yes — when the discriminant $(1-p_0)^2-4q_0$ is negative, the roots are a complex-conjugate pair $\alpha\pm i\beta$ . The real solutions are then $x^\alpha\cos(\beta\ln x)$ and $x^\alpha\sin(\beta\ln x)$ , which oscillate faster and faster as $x\to0$ because $\ln x\to-\infty$ . Physically this signals a scale-invariant instability: it appears in the fall of a quantum particle to the centre of an attractive $1/r^2$ potential (the " $-\alpha/r^2$ collapse"), where the infinitely oscillating wavefunction has no ground state until the singularity is regularized. So complex exponents are a red flag that the idealized problem is pathological and needs a short-distance cutoff.

If an integer root difference does not guarantee a logarithm, how do I actually check?

Compute the smaller root's recurrence up to the resonant step $n=N=r_1-r_2$ . At that step the prefactor $F(N+r_2)=F(r_1)$ is zero, so the equation reads $0\cdot a_N=(\text{a specific sum of earlier terms})$ . Evaluate that sum. If it is zero, the equation is $0=0$ , $a_N$ is free, and you get a pure Frobenius series with no logarithm. If it is nonzero, the equation is a contradiction, and the only way to restore consistency is to add $C\,y_1\ln x$ with $C\ne0$ . Bessel order $\tfrac12$ has a vanishing sum (no log); integer-order Bessel has a nonzero sum (log in $Y_n$ ).

Does every second-order ODE have an indicial equation at a given point?

Only at ordinary points and regular singular points. At an ordinary point the indicial roots are $0$ and $1$ (two Taylor series). At a regular singular point you get a genuine quadratic in $r$ . But at an irregular singular point — where $xp$ or $x^2q$ fails to be analytic, so $q$ has worse than a double pole — the Frobenius construction collapses: there is no finite leading exponent, and solutions typically carry essential singularities like $e^{1/x}$ . Those require asymptotic or WKB methods instead. The first thing to do with any singular point is therefore to test whether $xp$ and $x^2q$ are analytic; only if they are does an indicial equation even exist.