For so long I’ve been skeptical about the classic approach of the “Theory of Algorithms” and its misuse and misunderstanding by many software engineers and programmers. Big O notation, the Big Theta Θ and Big Omega Ω notations are often not useful for comparing the performance of algorithms in practice. They are often not even useful for classifying algorithms. They are useful for determining theoretical limits of an algorithms’ performance. In other words, their theoretical lower bound, upper bound or both.
I’ve had to painfully and carefully argue this point a few times as an interviewee and many times as part of a team of engineers. In the first case it can mean the difference between impressing the interviewer or missing out on a great career opportunity due simply to ignorance and/or incorrigibility of the person interviewing you. In the latter it could mean wasted months or even years in implementation effort and/or a financial failure in the worst case.
In practice the O-notation approach to algorithmic analysis can often be quite misleading. Quick Sort vs. Merge Sort is a great example. Quick Sort is classified as time quadratic O(n²) and Merge Sort as time log-linear O(n log n) according to O-notation. In practice however, Quick Sort often performs twice as fast as Merge Sort and is also far more space efficient. As many folks know this has to do with the typical inputs of these algorithms in practice. Most engineers I know would still argue that Merge Sort is a better solution and apparently Robert has had the same argumentative response even though he is an expert in the field. In the lecture he kindly says the following : “… Such people usually don’t program much and shouldn’t be recommending what practitioners do”.
There are many more numerous examples of where practical application does not align with the use of O-notation. Also, detailed analysis of algorithmic performance just takes too long to be useful in practice most of the time. So what other options do we have ?
There is a better way. An emerging science called “Analytic Combinatorics” pioneered by Robert Sedgewick and the late Philippe Flajolet over the past 30 years with the first (and only) text appearing in 2009 called Analytic Combinatorics. This approach is based on the scientific method and provides an accurate and more efficient way to determine the performance of algorithms(and classify them correctly). It even makes it possible to reason about an algorithms’ performance based on real-world input. It also allows for the generation of random data for a particular structure or structures, among other benefits.
For an introduction by the same authors there is An Introduction to the Analysis of Algorithms(or the free PDF version)and Sedgewick’s video course. Just to make it clear how important this new approach is going to be to computer science (and other sciences), here’s what another CS pioneer has to say :
“[Sedgewick and Flajolet] are not only worldwide leaders of the field, they also are masters of exposition. I am sure that every serious computer scientist will find this book rewarding in many ways.” —From the Foreword by Donald E. Knuth