If y1 and y2 are two solutions of dxdy+P(x)y=Q(x), and y2=y1u, then show that
u=1+ae−∫(y1Q)dx,
where a is a non-zero constant.
Technique
Substitute y2=y1u into the linear equation and use the fact that y1 already satisfies it to annihilate the P-terms; what survives is a separable first-order equation for u alone. Integrate, and read off why the constant cannot be zero.
Solution
Step 1 — State the hypotheses precisely (and the standing assumption).
Both functions satisfy the same linear equation on a common interval I:
dxdy1+P(x)y1=Q(x),(1)dxdy2+P(x)y2=Q(x).(2)
For the substitution y2=y1u to define u at all we need y1=0 on I; and for y2 to be a second (i.e. different) solution we need y2≡y1, so u≡1. Both are taken throughout.
Step 2 — Put y2=y1u into equation (2).
By the product rule, dxdy2=y1dxdu+udxdy1. Substituting in (2):
y1dxdu+udxdy1+Py1u=Q,
and grouping the two terms carrying u:
y1dxdu+u=Qby (1)(dxdy1+Py1)=Q.
This is the load-bearing step: the bracket is exactly the left side of (1), so it equals Q. Note that P has now disappeared completely. Hence
y1dxdu+uQ=Q⟹y1dxdu=Q(1−u).(3)
Step 3 — Separate and integrate.
Equation (3) is separable. Since u≡1 we may divide by 1−u (on any subinterval where 1−u=0; by uniqueness for the linear equation, u=1 at one point would force y2≡y1, so in fact 1−u never vanishes):
1−udu=y1Qdx.
Integrating both sides,
−ln∣1−u∣=∫y1Qdx+c,
so
∣1−u∣=e−ce−∫(Q/y1)dx⟹1−u=Ae−∫(Q/y1)dx,
where A=±e−c is an arbitrary non-zero constant (the sign is absorbed on removing the modulus).
Therefore
u=1−Ae−∫(Q/y1)dx=1+ae−∫(Q/y1)dx,a:=−A.
Step 4 — Why a=0.
If a=0 the formula gives u≡1, hence y2=y1u=y1. But y1 and y2 were taken to be two solutions, i.e. distinct ones; a repeat of y1 carries no new information and the substitution y2=y1u would be vacuous. Hence a=0. (Equivalently: a=0 is exactly the degenerate case excluded by u≡1 in Step 2, which is what allowed the division by 1−u.)
Step 5 — Consistency check (worth one line in the exam).
Subtracting (1) from (2), w:=y2−y1 satisfies the homogeneous equation w′+Pw=0, so w=Ce−∫Pdx. Our answer gives
w=y1(u−1)=ay1e−∫(Q/y1)dx.
The two agree because, dividing (1) by y1,
y1y1′−y1Q=−P⟹dxd[lny1−∫y1Qdx]=−P,
i.e. y1e−∫(Q/y1)dx=elny1−∫(Q/y1)dx is precisely e−∫Pdx up to a constant. The result is therefore internally consistent with the standard structure “general solution = particular + homogeneous”.