The equivalence of folding left and right over Peano numbers and lists makes it possible to minimalistically inter-derive (1) structurally recursive functions in direct style, (2) structurally tail-recursive functions that use an accumulator, and (3) structurally tail-recursive functions in delimited continuation-passing style, using Ohori and Sasano’s lightweight fusion by fixed-point promotion. When the fold-left and the fold-right functions account for primitive iteration for Peano numbers, this equivalence is unconditional. When they account for primitive recursion for Peano numbers, this equivalence is modulo left permutativity of their induction-step parameter – a property which is more general than associativity and commutativity. And when they account for primitive iteration or for primitive recursion over lists, this equivalence is modulo left permutativity of their induction-step parameter if these two fold functions have the same type. Since the 1980s, however, the two fold functions for lists do not have the same type: the arguments for their induction-step parameter are swapped, a re-ordering that complicated Bird and Wadler’s duality theorems and whose history is reviewed in an appendix. Without this re-ordering, Bird and Wadler’s second duality theorem more visibly accounts for “re-bracketing,” which is a key step to make recursive programs tail recursive in the general area of program development, from Cooper in the 1960s and onwards.

The Garsia–Wachs algorithm is an algorithm for building a binary leaf tree whose cost is as small as possible. The problem and the algorithm are described in more detail below, but the task is essentially the same as that of building a Huffman coding tree with the added constraint that the fringe of the tree has to be exactly the given list of inputs (in Huffman coding, the fringe of the tree can be any permutation of the input). As we will show below, the Garsia–Wachs algorithm can be implemented with a linearithmic running time—a running time of O (n log n) steps for an input of length n, the same time bound as for Huffman coding.

The Knuth–Morris–Pratt (KMP) algorithm for string search is notoriously difficult to understand. Lost in a sea of index arithmetic, most explanations of KMP obscure its essence. This paper constructs KMP incrementally, using pictures to illustrate each step. The end result is easier to comprehend. Additionally, the derivation uses only elementary functional programming techniques.

Fenwick trees, also known as binary indexed trees are a clever solution to the problem of maintaining a sequence of values while allowing both updates and range queries in sublinear time. Their implementation is concise and efficient—but also somewhat baffling, consisting largely of nonobvious bitwise operations on indices. We begin with segment trees, a much more straightforward, easy-to-verify, purely functional solution to the problem, and use equational reasoning to explain the implementation of Fenwick trees as an optimized variant, making use of a Haskell EDSL for operations on infinite two’s complement binary numbers.

The setting is a class on functional programming. There are four students: Anne, Jack, Mary and Theo.

Functional programmers have many things for which to thank the late David Turner: design decisions he made in his languages SASL, KRC, and Miranda over the last 50 years are still influential and inspirational now. In particular, Turner was a strong advocate of lazy evaluation and of list comprehensions. As an illustration of these techniques, he popularized a one-line recursive “sieve” to generate the infinite list of prime numbers.

Turner called this algorithm The Sieve of Eratosthenes. In a lovely paper called “The Genuine Sieve of Eratosthenes”, Melissa O’Neill argued that Turner’s program is not in fact a faithful implementation of the algorithm, and gave a detailed presentation using priority queues of the real thing. She included a variation by Richard Bird, which uses only lists but makes clever use of circular programming. Bird describes his circular program again in his textbook “Thinking Functionally with Haskell”, and sets its proof of correctness as an exercise. In particular, why is this circular program productive? Unfortunately, Bird’s hint for a solution is incorrect. So what should a proof look like?

One of the last projects Turner worked on was the notion of “Total Functional Programming”. He observed that most programs are already structurally recursive or corecursive, therefore guaranteed respectively terminating or productive; he conjectured that “with more practice we will find this is always true”. We explore Bird’s circular Sieve of Eratosthenes as a challenge problem for Turner’s Total Functional Programming.

From the outset, lambda calculus represented natural numbers through iterated application. The successor hence adds one more application, and the predecessor removes. In effect, the predecessor un-applies a term—which seemed impossible, even to Church. It took Kleene a rather oblique glance to sight a related representation of numbers, with an easier predecessor. Let us see what we can do if we look at this old problem with today’s eyes. We discern the systematic ways to derive more predecessors—smaller, faster, and sharper—while keeping all teeth.

Vectors—or length-indexed lists—are classic example of a dependent type. Yet, most tutorials stay clear of any function on vectors whose definition requires non-trivial equalities between natural numbers to type check. This pearl shows how to write functions, such as vector reverse, that rely on monoidal equalities to be type correct without having to write any additional proofs. These techniques can be applied to many other functions over types indexed by a monoid, written using an accumulating parameter, and even be used to decide arbitrary equalities over monoids ‘for free.’

This pearl defines a translation from well-typed lambda terms to combinatory logic, where both the preservation of types and the correctness of the translation are enforced statically.

The other day, I was assembling lecture material for a course on Agda. Pursuing an application-driven approach, I was looking for correctness proofs of popular algorithms. One of my all-time favourites is Huffman data compression (Huffman, 1952). Even though it is probably safe to assume that you are familiar with this algorithmic gem, a brief reminder of the essential idea may not be amiss.

Let’s say we want to create the two charts in Figure 1. The chart on the left is a bar chart that shows two different values for each bar. The chart on the right consists of two line charts that share the x axis with parts of the timeline highlighted using two different colors.

Comparison networks offer an attractive framework for studying parallel sorting algorithms. The visual presentation of such hard-wired networks makes processes easy to conceptualize without the overhead of programming syntax or the semantic complications of concurrency. Though simple in appearance they have a surprisingly rich structure. In this pearl we delve a little into their theory.

Some top-down problem specifications, if executed, may compute sub-problems repeatedly. Instead, we may want a bottom-up algorithm that stores solutions of sub-problems in a table to be reused. How the table can be represented and efficiently maintained, however, can be tricky. We study a special case: computing a function ${\mathit{h}}$ taking lists as inputs such that ${\mathit{h}\;\mathit{xs}}$ is defined in terms of all immediate sublists of ${\mathit{xs}}$. Richard Bird studied this problem in 2008 and presented a concise but cryptic algorithm without much explanation. We give this algorithm a proper derivation and discovered a key property that allows it to work. The algorithm builds trees that have certain shapes—the sizes along the left spine is a prefix of a diagonal in Pascal’s triangle. The crucial function we derive transforms one diagonal to the next.

We consider the task of constructing 2-3 trees. Given a sequence of elements we seek to build a 2-3 tree–in linear time–that contains the elements in symmetric order. We discuss three approaches: top-down, bottom-up, and incremental. The incremental approach is more flexible than the other two in that it allows us to interleave the construction work with other operations, for example, queries.

Consider the following puzzle: given a number, remove k digits such that the resulting number is as large as possible. Various techniques are employed to derive a linear-time solution to the puzzle: we justify the structure of a greedy algorithm by predicate logic, give a constructive proof of the greedy condition using a dependently typed proof assistant and calculate the greedy step as well as the final, linear-time optimisation by equational reasoning.