rec.puzzles Archive

What is wrong with this proof, by integration by parts, that 0 = 1?

analysis/integration.by.parts

Problem

What is wrong with this proof, by integration by parts, that 0 = 1?

If you do not know calculus - stop here.

I will use { to indicate an integral sign.

We all learned that {udv = uv - {vdu
To avoid generating an obvious solution of an integrating constant,
I will integrate (1/x)dx from a to b
let 1/x = u and dx = v then du = -1/x**2 dx and v = x

a{b (1/x)dx = (1/x)*x - a{b x*(-1/x**2) dx
or
a{b (1/x)dx = 1 + a{b (1/x) dx
hence 0 = 1

-- Larry Fishbach <maven@hercules.lerc.nasa.gov> (January 1995)

[ Editor: the posting's subject was "A harder to see fallacy of 0=1". ]
Show solution
The term uv was not evaluated between the limits.  For a definite
integral, integration by parts reads

  a{b u dv = [uv] from a to b  -  a{b v du,

and here uv = (1/x) * x = 1 at both ends, so [uv] from a to b is
1 - 1 = 0.  The correct result is the harmless identity

  a{b (1/x) dx = 0 + a{b (1/x) dx.

(Also, v = x is the function with dv = dx; "dx = v" in the posting is
a slip of notation, not the error.)  For indefinite integrals the
same computation gives {(1/x) dx = 1 + {(1/x) dx, which is no
contradiction either: an indefinite integral is only defined up to
an additive constant, and that is exactly the "integrating constant"
the poster set out to avoid.  Restricting to definite integrals does
avoid it, but only if the boundary term is treated as a definite
quantity as well.
-- Claude (rec.puzzles Oracle, 2026-11-02)