4.6.18Ordinary Differential Equations

Frobenius method — regular singular points

2,044 words9 min readdifficulty · medium2 backlinks

A way to solve y+P(x)y+Q(x)y=0y'' + P(x)y' + Q(x)y = 0 near a point where the simple power-series method fails.

Why do we even need this?

So the WHAT: we look for y=xrn0anxny = x^r\sum_{n\ge0} a_n x^n where rr is to be determined, not assumed integer.


Classifying the singular point (the gatekeeper)

Write the ODE in standard form: y+P(x)y+Q(x)y=0.y'' + P(x)\,y' + Q(x)\,y = 0.

WHY these exact powers? Because if PP blows up no worse than 1/(xx0)1/(x-x_0) and QQ no worse than 1/(xx0)21/(x-x_0)^2, then multiplying by xrx^r exactly balances the worst singularity in the equation. Stronger blow-ups can't be balanced by a single power — that's why irregular points break the method.

Figure — Frobenius method — regular singular points

Deriving the indicial equation FROM SCRATCH

Take x0=0x_0=0 and write the RSP equation in the cleanest form by multiplying by x2x^2: x2y+xp(x)y+q(x)y=0,x^2 y'' + x\,p(x)\,y' + q(x)\,y = 0, where p(x)=k0pkxkp(x)=\sum_{k\ge0}p_k x^k and q(x)=k0qkxkq(x)=\sum_{k\ge0}q_k x^k are analytic.

Step 1 — Assume the Frobenius ansatz. y=n=0anxn+r,a00.y = \sum_{n=0}^{\infty} a_n x^{n+r}, \quad a_0\neq 0. Why a00a_0\neq0? We demand the lowest power present is xrx^r; if a0=0a_0=0 we just relabel rr, so insist it's nonzero.

Step 2 — Differentiate term by term. y=nan(n+r)xn+r1,y=nan(n+r)(n+r1)xn+r2.y' = \sum_n a_n (n+r)x^{n+r-1}, \qquad y'' = \sum_n a_n (n+r)(n+r-1)x^{n+r-2}.

Step 3 — Plug in and look ONLY at the lowest power xrx^r (i.e. n=0n=0, pp0p\to p_0, qq0q\to q_0): x2ya0r(r1)xr,xpya0p0rxr,qya0q0xr.x^2 y'' \to a_0 r(r-1)x^r,\quad x p\, y' \to a_0 p_0 r\, x^r,\quad q\,y \to a_0 q_0 x^r. Adding and setting the coefficient of xrx^r to zero (with a00a_0\neq0):

Step 4 — Recurrence for the rest. Setting the coefficient of xn+rx^{n+r} to zero gives [(n+r)(n+r1)+p0(n+r)+q0]an=k=1n[pk(nk+r)+qk]ank.\big[(n+r)(n+r-1)+p_0(n+r)+q_0\big]a_n = -\sum_{k=1}^{n}\big[p_k(n-k+r)+q_k\big]a_{n-k}. The bracket on the left is F(n+r)F(n+r) where F(s)=s(s1)+p0s+q0F(s)=s(s-1)+p_0 s+q_0 is the indicial polynomial. So ana_n is solvable as long as F(n+r)0F(n+r)\neq0.


The three cases (WHY the second solution changes form)

Let r1r2=r_1 - r_2 = the root difference.

WHY a log in Cases 2 & 3? When F(n+r2)=0F(n+r_2)=0 at some nn, the recurrence demands division by zero. Reduction of order y2=v(x)y1y_2=v(x)y_1 then produces a dx/x=lnx\int dx/x = \ln x term. The log is the mathematics' way of supplying a second independent solution when the power series ran out of room.


Worked Example 1 — get p0,q0p_0,q_0 and solve the indicial equation

ODE:   2x2y+3xy(1+x)y=0.\;2x^2 y'' + 3x\,y' - (1+x)y = 0.

Step 1: Standard form. Divide by 2x22x^2: y+32xy1+x2x2y=0.y'' + \frac{3}{2x}y' - \frac{1+x}{2x^2}y=0. Why? We need P,QP,Q to read off the singularity.

Step 2: Check RSP. xP=32xP = \tfrac32 (analytic), x2Q=1+x2x^2Q = -\tfrac{1+x}{2} (analytic). ✔ So x=0x=0 is a regular singular point. Why this step? Frobenius is only guaranteed at an RSP.

Step 3: Read p0,q0p_0,q_0. p0=32p_0 = \tfrac32, q0=x2Qx=0=12q_0 = x^2Q|_{x=0} = -\tfrac12.

Step 4: Indicial equation. r(r1)+32r12=r2+12r12=02r2+r1=0(2r1)(r+1)=0.r(r-1)+\tfrac32 r -\tfrac12 = r^2+\tfrac12 r -\tfrac12=0 \Rightarrow 2r^2+r-1=0 \Rightarrow (2r-1)(r+1)=0. Roots r1=12,  r2=1r_1=\tfrac12,\; r_2=-1. Difference =32Z=\tfrac32\notin\mathbb ZCase 1, two clean Frobenius series. Why this matters? We instantly know NO logarithm appears.


Worked Example 2 — full recurrence (Bessel-type, equal roots)

ODE:   x2y+xy+x2y=0\;x^2y'' + xy' + x^2 y = 0 (Bessel order 0).

Step 1: P=1/x, Q=1P=1/x,\ Q=1. xP=1xP=1, x2Q=x2x^2Q=x^2, both analytic → RSP. p0=1, q0=0p_0=1,\ q_0=0.

Step 2: Indicial. r(r1)+r+0=r2=0r1=r2=0r(r-1)+r+0=r^2=0\Rightarrow r_1=r_2=0 (Case 2 — expect a log in y2y_2). Why this step? Equal roots predict the structure before we compute anything.

Step 3: Plug y=anxny=\sum a_n x^{n} (since r=0r=0). Then ann(n1)xn+annxn+anxn+2=0.\sum a_n n(n-1)x^n+\sum a_n n x^n + \sum a_n x^{n+2}=0. Combine first two: ann2xn+anxn+2=0\sum a_n n^2 x^n + \sum a_n x^{n+2}=0.

Step 4: Match powers xmx^m. amm2+am2=0am=am2m2.a_m m^2 + a_{m-2}=0\Rightarrow a_m = -\dfrac{a_{m-2}}{m^2}. Why this step? Shifting index nm2n\to m-2 aligns powers so we can equate coefficients.

Step 5: Iterate. a1=0a_1=0 (since 12a1=01^2 a_1=0), so all odd terms vanish. a2=a04,a4=a216=a064,,a2k=(1)ka022k(k!)2.a_2=-\frac{a_0}{4},\quad a_4=-\frac{a_2}{16}=\frac{a_0}{64},\dots,\quad a_{2k}=\frac{(-1)^k a_0}{2^{2k}(k!)^2}. With a0=1a_0=1 this is J0(x)=k(1)k(k!)2(x/2)2kJ_0(x)=\sum_{k}\frac{(-1)^k}{(k!)^2}(x/2)^{2k} — the first Bessel function. The second solution Y0Y_0 carries the predicted lnx\ln x.


Common mistakes (Steel-manned)


Active Recall

Recall Quick self-test
  • What two functions must be analytic for x0x_0 to be an RSP?
  • Write the indicial equation and define p0,q0p_0,q_0.
  • Which case forces a logarithm with certainty?
  • In Example 1, why was there NO log?
Recall Feynman: explain to a 12-year-old

Imagine a swing whose chain is frayed right at the top pivot (x=0x=0). A normal smooth wiggle-formula can't describe motion at a frayed spot. So we cheat: we say "the motion looks like xrx^r times a smooth wiggle," and we let the equation itself tell us the magic exponent rr by solving a tiny quadratic (the indicial equation). Sometimes both magic numbers work nicely; sometimes they're the same or differ by a whole step, and then a "lnx\ln x" sneaks in to give us a genuinely different second motion.


Flashcards

What makes x0x_0 a regular singular point?
p(x)=(xx0)Pp(x)=(x-x_0)P and q(x)=(xx0)2Qq(x)=(x-x_0)^2Q are both analytic at x0x_0.
Frobenius ansatz about x=0x=0?
y=n0anxn+ry=\sum_{n\ge0}a_n x^{n+r} with a00a_0\neq0.
Indicial equation?
r(r1)+p0r+q0=0r(r-1)+p_0 r+q_0=0, with p0=limx0xPp_0=\lim_{x\to0}xP, q0=limx0x2Qq_0=\lim_{x\to0}x^2Q.
Why must a00a_0\neq0?
So xrx^r is genuinely the lowest power; otherwise we'd just redefine rr.
Case with equal indicial roots?
y2=y1lnx+xrn1cnxny_2=y_1\ln x + x^{r}\sum_{n\ge1}c_n x^n — log forced.
Roots differ by positive integer NN — general y2y_2?
y2=κy1lnx+xr2cnxny_2=\kappa y_1\ln x + x^{r_2}\sum c_n x^n, κ\kappa possibly 0.
Why can a log appear?
The recurrence hits F(n+r)=0F(n+r)=0 (division by zero); reduction of order yields dx/x=lnx\int dx/x=\ln x.
When does NO log appear with distinct roots?
When r1r2Zr_1-r_2\notin\mathbb Z (Case 1).
What does irregular singular point mean for Frobenius?
Method not guaranteed; the standard series may fail entirely.
Indicial roots of x2y+xy+x2y=0x^2y''+xy'+x^2y=0?
r=0r=0 (double) — gives J0J_0 and a ln\ln-carrying Y0Y_0.

Connections

Concept Map

motivates

uses ansatz

r not assumed integer

classify point

p = x P and q = x^2 Q analytic

blow-up too strong

method may fail

multiply by x^2 and sub ansatz

coefficient set to zero

r r-1 + p0 r + q0 = 0

determine structure

higher powers

generates coefficients

Power series method fails at singular pts

Frobenius method

y = x^r sum a_n x^n

Captures singular behaviour

Standard form y'' + Py' + Qy = 0

Regular singular point?

RSP valid for Frobenius

Irregular singular pt

Lowest power x^r term

Indicial equation

Indicial roots r1 r2

Solution form

Recurrence for a_n

Hinglish (regional understanding)

Intuition Hinglish mein samjho

Dekho, jab hum y+Py+Qy=0y''+Py'+Qy=0 ko solve karte hain, normally hum y=anxny=\sum a_n x^n (simple power series) maan lete hain. Lekin kuch equations (jaise Bessel, Legendre) me x=0x=0 par PP ya QQ "phat" jaate hain — yaani analytic nahi rehte. Aise point ko singular point kehte hain. Agar sirf xPxP aur x2Qx^2Q analytic ho jaayein (yaani singularity bahut zyada strong na ho), toh use regular singular point kehte hain, aur yahin Frobenius method kaam karta hai.

Frobenius ka jugaad simple hai: hum y=xranxny=x^r\sum a_n x^n maante hain — yaani normal series ke saath ek extra unknown power xrx^r multiply kar dete hain. Yeh rr humein khud equation se nikalna padta hai. Lowest power xrx^r ka coefficient zero karke milta hai indicial equation: r(r1)+p0r+q0=0r(r-1)+p_0 r+q_0=0, jahan p0=limxPp_0=\lim xP aur q0=limx2Qq_0=\lim x^2Q. Iske do roots r1,r2r_1, r_2 pura structure decide karte hain.

Ab teen cases: agar roots ka difference integer nahi hai, toh do clean series mil jaati hain — easy. Agar roots barabar hain, toh sirf ek exponent hai, isliye doosra solution majboori me lnx\ln x leke aata hai. Aur agar roots integer se differ karte hain, toh log aa bhi sakta hai aur nahi bhi (coefficient κ\kappa zero ho sakta hai). Log isliye aata hai kyunki recurrence me kahin "zero se divide" ka problem aata hai, aur reduction of order se dx/x=lnx\int dx/x=\ln x nikal aata hai.

Yaad rakhna: indicial equation me P,QP,Q directly mat daalna (woh toh infinity ho jaate hain) — pehle p0,q0p_0,q_0 ko limit se nikaalo. Aur hamesha pehle RSP check karo, kyunki irregular point par Frobenius guarantee nahi karta. Bas itna samajh gaye toh Bessel aur Legendre dono aaram se nikal jaayenge.

Go deeper — visual, from zero

Test yourself — Ordinary Differential Equations

Connections