.123456... = x + 2 x^2 + 3 x^3 + ... with x = 1/10.
Then you have (x + 2 x^2 + 3 x^3 + ...) = (x + x^2 + x^3 + x^4 + ...) + (x^2 + x^3 + x^4 + x^5 + ...) + (x^3 + x^4 + x^5 + x^6 + ...) (count the number of occurrences of each power of x^n on the right-hand side)
and from the sum of a geometric series the RHS is x/(1-x) + x^2/(1-x) + x^3/(1-x) + ..., which itself is a geometric series and works out to x/(1-x)^2. Then put in x = 1/10 to get 10/81.
Now 0.987654... = 1 - 0.012345... = 1 - (1/10) (10/81) = 1 - 1/81 = 80/81.
Don't need the clutter of infinite series and polynomials:
Isn't it essentially the same thing, but less formal
0.1111... is just a notation for (x + x^2 + x^3 + x^4 + ...) with x = 1/10
1/9 = 0.1111... is a direct application of the x/(1-x) formula
The sum of 0.0111... + 0.00111... ... = 0.012345... part is the same as the "(x + 2 x^2 + 3 x^3 + ...) = (x + x^2 + x^3 + x^4 + ...) + (x^2 + x^3 + x^4 + x^5 + ...)" part (but divided by 10)
And 1/81 = 1/9 * 1/9 ... part is the x/(1-x)^2 result
This is better than my answer, at least if you can get your brain to interpret it in base b. In that case the first two lines would become
I don't know who downvoted this, but it's correct.
The use of series is a little "sloppy", but x + 2 x^2 + 3 x^3 + ... has absolute uniform convergence when |x|<r<1, even more importantly that it's true even for complex numbers |z|<r<1.
The super nice property of complex analysis is that you can be almost ridiculously "sloppy" inside that open circle and the Conway book will tell you everything is ok.
[I'll post a similar proof, but mine use -1/10 and rounding, so mine is probably worse.]