Deterministic
Definition: A sequence of numbers $X_n$ is said to be $O(1)$ if there exist $M$ and $n_0$ such that
\[\abs{X_n} < M\]for all $n>n_0$. Likewise, $X_n$ is said to be $O(r_n)$ if there exist $M$ and $n_0$ such that for all $n>n_0$,
\[\abs*{\frac{X_n}{r_n}} < M.\]Definition: A sequence of numbers $X_n$ is said to be $o(1)$ if it converges to zero. Likewise, $X_n$ is said to be $o(r_n)$ if
\[\frac{X_n}{r_n} \to 0\]as $n \to \infty$.
Stochastic
Definition: A sequence of random vectors $\x_n$ is said to be $o_p(1)$ if it converges to $\zero$ in probability. Furthermore, $\x_n$ is said to be $o_p(r_n)$ if
\[\frac{\x_n}{r_n} \inP \zero.\]Definition: A sequence of random vectors $\x_n$ is said to be $O_p(1)$ if it is bounded in probability. Furthermore, $\x_n$ is said to be $O_p(r_n)$ if $\x_n/r_n$ is bounded in probability.
Algebra of $O, o$ notation
The following rules apply to both deterministic and stochastic representations (here, $a \le b$):
\[\begin{alignat*}{2} O(1) + O(1) &= O(1) & O\{O(1)\} &= O(1) \\ o(1) + o(1) &= o(1) & o\{O(1)\} &= o(1) \\ o(1) + O(1) &= O(1) & o(r_n) &= r_no(1) \\ O(1)O(1) &= O(1) & O(r_n) &= r_nO(1) \\ O(1)o(1) &= o(1) & O(n^a) + O(n^b) &= O(n^b) \\ \{1+o(1)\}^{-1} &= O(1) &\qquad\qquad o(n^a) + o(n^b) &= o(n^b) \end{alignat*}\]Source: Asymptotic statistics (1998), van der Vaart A. Cambridge.
Proof: This is an example proof, showing that $o(1) + O(1) = O(1)$. Other results can be derived with a similar technique.
Let $A_n = o(1)$, $B_n = O(1)$, and $\eps > 0$.
\[\begin{alignat*}{2} \tag*{$\tcirc{1}$} \exists N_a: n > N_a \implies \abs{A_n} &< \eps &\hspace{4em}& A_n=o(1) \\ \tag*{$\tcirc{2}$} \exists N_b, M: n > N_b \implies \abs{B_n} &< M && B_n=O(1) \\ n > \max(N_a, N_b) \implies \abs{A_n + B_n} &\le \abs{A_n} + \abs{B_n} && \href{norm.html}{\text{Triangle inequality}} \\ &< \eps + \abs{B_n} && \tcirc{1} \\ &< \eps + M && \tcirc{2} \end{alignat*}\]