# What does it mean to add up O-terms with different variables? [duplicate]

Is this true? O(n) + O(k) =O(n+k).I have searched for it ,the answers were quite ambiguous and I couldn't find a good explanation.

## marked as duplicate by Evil, adrianN, fade2black, Yuval Filmus, David RicherbyOct 28 '17 at 16:35

Yes, this is true.

The LHS means that for some $N$, $M$ and some constants $c_n,c_k$,

$$n\ge N,k\ge M\implies f(n,k)\le c_nn+c_kk.$$

The RHS means that for some $N'$, $M'$ and some constant $c$,

$$n\ge N',k\ge M'\implies f(n,k)\le c(n+k).$$

These two statements are equivalent if you take $c\leftarrow\max(c_n,c_k),N'\leftarrow N,M'\leftarrow M$ one way and $c_n=c_k\leftarrow c,N\leftarrow N',M \leftarrow M'$ the other.

• Note for reference: it's not at all clear that the RHS should mean that. – Raphael Oct 27 '17 at 10:42
• Referring to my answer, this pertains to reading the original expression as $O_{n,k}(n) + O_{n,k}(k) = O_{n,k}(n + k)$. This is probably the intended interpretation, e.g. if $n$ and $k$ hold some meaning in the context in which the expression appears. In isolation, it's an ambiguous and hence rather vacuous thing to write down; but that's of course neither prithvi parre's nor Yves Daoust's fault. – Raphael Oct 27 '17 at 10:54

Without further context, the equation

$\qquad O(n) + O(k) = O(n+k)$

is ambiguous and therefore rather meaningless. I'll explain why.

First, recall that Landau notation defines sets. The definition commonly used in CS is, with $f : \mathbb{N} \to \mathbb{R}_{>0}$,

$\qquad O(f) = \{ g : \mathbb{N} \to \mathbb{R}_{>0} \mid \exists\, c > 0, n_0 \geq 0.\ \forall\, n \geq n_0.\ g(n) \leq cf(n) \}$.

Note how $n$ is a bound variable inside the set expression. It's only a label for the abstract parameter of the mathematical functions. Therefore,

$\qquad O(n) = O(k)$.

Note how, with this definition, it's completely unclear what $O(n + k)$ is supposed to mean. That is not (only) an oversight in the definition; as referenced here and here, there simply is no rigorous definition that does all we want. The use of multi-parameter $O$ in the literature is usually sloppy, if not to say wrong, and only to be read in a hand-waving kind of way.

Now, in algorithm analysis such terms do not appear on paper on their own. Say, $n$ is the number of nodes of the input graph, and $k$ its diameter. Then, we may perform two subroutines, one that runs in time $O(n)$ (it doesn't depend on the diameter), and the other runs in time $O(k)$ (it doesn't depend on the number of nodes). Then, we sort of understand what $O(n + k)$ is supposed to mean.

Now, back to your original expression. The question is, what's variable and what's constant? To make the point, consider $O(n + k^2)$ instead. We can read it as

• $O(n)$ if $k$ is constant,
• $O(k^2)$ if $n$ is constant, and
• unsimplified (with unclear meaning) if neither is constant.

To make this explicit, I write $O_x$ to imply that $x$ is the variable used in the definition of $O$ here. In the above example, we get

• $O_n(n + k^2) = O_n(n) = O_x(x)$,
• $O_k(n+k^2) = O_k(k^2) = O_x(x^2)$, and
• $O_{n,k}(n + k^2) =\ ?$.

Here we immediately see that the three are different things!

Now we have notation to write down what we really mean in our setting:

$\qquad O_{n,k}(n) + O_{n,k}(k) = O_{n,k}(n + k)$.

Here, $n$ and $k$ are both independent variables!
(Let's ignore for a moment that $O_{x, y}$ is not defined by the definition I give at the top; it's inherently single-parameter. We're hand-waving, as I said.)
Note how this is different from

$\quad O_n(n) + O_k(k) \neq O_{n,k}(n+k)$,

which is what just inserting the definiton of $O$ in the straight-forward way would give you.

While the "result" read as a two-variable $O$-term doesn't have a lot of meaning on its own, we can use it as a stepping stone! Once we fix a functional relationship between $n$ and $k$ -- we may investigate $k \sim n$ vs $n \sim \log n$ vs $k = 10$, for instance -- we get back to single-variable $O$ which is perfectly well-defined.

This is not a mathematically rigorous approach, but if we don't accidentally "simplify" away terms in the hand-wavy two-parameter world that we shouldn't, inserting the functional relationship afterwards gives the same result as doing the whole analysis with it in mind -- it's just a substituion. Some care has to be taken.