I'm not sure I agree with this. Conceptually, if you simply think of the indefinite integral as not yielding a function, but rather as relating a function to an equivalence class of functions (the equivalence being that they differentiate to the same function), with +, -, etc. overloaded accordingly to such equivalence classes, then you don't have to worry about constants in your formulation of the rules at all, until you want to remove the $$\int\,\cdot\, dx$$ anti-differentiation operator. And then, a simple, single $$+C$$ (to indicate the general case) will do. Voilá, no mucking about with different "constants".
Right, and this is exactly what they taught me back in introductory calculus. But there's some subtlety to the question of adding in that one constant - the "naive" rule that people seem to use there is that you only add a constant when explicitly evaluating an indefinite integral. That's precisely the method through which mishin05's earlier logic produced the expression $$\int 0 ~dx = 0$$.
Which is of course completely valid if both sides of that equation are understood to refer to equivalence classes (i.e., there's still an implied final step of adding a constant to get the final answer, once all manipulations are finished). But it does not get understood that way by the unfamiliar, since they expect to see a $$+C$$ wherever an antiderivative is written out (according to the convention). The inclusion of a $$+C$$
is the notation for an equivalence class, after all, so it seems backwards to suppress that notation when what you want to emphasize is the fact the you're dealing in equivalence classes.
So the correct rule here is that you have to inspect the final expression and add in whatever constant may be required for it to make sense as an equivalence class relation. Of course, those of us who've already mastered the subject aren't bothered by that, but it seems too much to ask of those who haven't (which is effectively the entire audience for indefinite integrals, given how rarely they're used outside of introductory calculus). Rather, a useful convention should guarantee that following concrete rules will produce expressions with all of the appropriate constants in the appropriate places. Otherwise, the convention can't be used to identify and locate errors in inferences that employ it, which is a huge handicap for anyone trying to learn the subject.
These problems are all products of attempts to suppress notation for reasons of aesthetics/shorthand, which seems like a mismatched approach for a subject that is almost entirely the domain of newcomers.
Said another way:
JamesR said:
When we write:
$$\int f(x)~dx = F(x)$$
it is understood that F(x) is any antiderivative of f(x). Since antiderivatives can differ only by a constant, the constant is implied. If we then write F(x) explicitly, we need to add the constant.
... it is my contention that such aspects should not be merely "understood" or "implied," but actually
notated as such to avoid the evident confusion. I.e., we're mixing and matching regular functions and variables with equivalences classes of functions here, but without any consistent, explicit difference in notations. So when mishin05 arrives at a result like $$\int 0 ~dx = 0$$, it is not apparent to him that this is an equivalence class relation, and not an assertion that the antiderivative of the zero function is exactly the zero function.
If the expectation is that the conventions should always produce an appropriate $$+C$$ when an antiderivative is written out (which seems common), then we need to ensure that all formulations conform to that. The linearity rule can be patched by adding an explicit $$+C$$, but it's my contention that always including the $$+C$$ everywhere is guaranteed to avoid all such issues (which surely there must be some other examples of, apart from the linearity rule?). Why not just denote equivalence classes explicitly all the time, especially given the audience?
JamesR said:
I don't think the additivity rule needs any modification
On second thought, the defect in the additivity rule is a result of the defect in the linearity rule (from passing the -1 through the indefinite integral), so if you patch the linearity rule you don't have a problem here.
But this does beg the question of how one is to write the results of expressions like $$\int f(x)~dx - \int f(x)~dx$$. The desired final expression would be $$C$$, but the "naive" conventions only require us to add a constant when "evaluating" an indefinite integral - it bears emphasizing that "removing" an antiderivative operator includes algebraic cancellation as well as explicit evaluation. Supposing we're to avoid a convention with a final step along the lines of "add in whatever constant may be required to get the right answer," that is.
JamesR said:
As that page looks right now (and I assume you're not the sole author), it looks overly complicated with all the explicit constants everywhere.
Well, part of my contention here is that indefinite integrals are a pedagogical mess and should be mostly ignored once one has completed introductory calculus, after all.
JamesR said:
As I say, my hunch is that careful definition of what we mean by the indefinite integral notation solves most of the problems.
The lazy formulation handles most of the cases just fine to begin with. The concern here is getting it to work
all the time (which I consider a reasonable expectation for math notation conventions), in order to patch the leaks that cranks like to squeeze through. Moreover I don't see much value in suppressing the notation here - given the audience, explicit notation is preferable (however ugly it may be).
Or, how about a compromise approach: by default, suppress the constants everywhere and simply deal in equivalence classes. If in doubt (or needing to demonstrate to someone not comfortable with equivalence classes), include the constants everywhere (including the indefinite integrals) to be totally explicit.