CHAPTER 8
Recursion
This chapter introduces the concept of recursion and recursive algorithms. First, the
definition of recursion and fundamental rules for recursive algorithms will be explored.
In addition, methods of analyzing efficiencies of recursive functions will be covered in
detail using mathematical notations. Finally, the chapter exercises will help solidify this
information.
Introducing Recursion
In math, linguistics, and art, recursion refers to the occurrence of a thing defined in
terms of itself. In computer science, a recursive function is a function that calls itself.
Recursive functions are often elegant and solve complex problems through the “divide-
and-conquer” method. Recursion is important because you will see it again and again
in the implementation of various data structures. Figure 8-1 shows a visual illustration of
recursion where the picture has smaller pictures of itself.
Figure 8-1. Recursion illustrated
99
Chapter 8 reCursion
Rules of Recursion
When recursive functions are implemented incorrectly, it causes fatal issues because the
program will get stuck and not terminate. Infinite recursive calls result in stack overflow.
Stack overflow is when the maximum number of call stacks of the program exceeds the
limited amount of address space (memory).
For recursive functions to be implemented correctly, they must follow certain rules
so that stack overflow is avoided. These rules are covered next.
Base Case
In recursion, there must be a base case (also referred to as terminating case). Because
recursive methods call themselves, they will never stop unless this base case is reached.
Stack overflow from recursion is most likely the result of not having a proper base case.
In the base case, there are no recursive function calls.
Let’s examine the following function, which prints numbers counting down from n to
0 as an example:
1 function countDownToZero(n) {
2 // base case. Stop at 0
3 if (n < 0) {
4 return; // stop the function
5 } else {
6 [Link](n);
7 countDownToZero(n - 1); // count down 1
8 }
9 }
10 countDownToZero(12);
The base case for this function is when n is smaller or equal to 0. This is because the
desired outcome was to stop counting at 0. If a negative number is given as the input,
it will not print that number because of the base case. In addition to a base case, this
recursive function also exhibits the divide-and-conquer method.
100
Chapter 8 reCursion
Divide-and-Conquer Method
In computer science, the divide-and-conquer method is when a problem is solved by
solving all of its smaller components. With the countdown example, counting down from
2 can be solved by printing 2 and then counting down from 1. Here, counting down from 1
is the part solved by “dividing and conquering.” It is necessary to make the problem
smaller to reach the base case. Otherwise, if the recursive call does not converge to a
base case, a stack overflow occurs.
Let’s now examine a more complex recursive function known as the Fibonacci
sequence.
Classic Example: Fibonacci Sequence
The Fibonacci sequence is a list of infinite numbers, each of which is the sum of the past
two terms (starting with 1).
1, 1, 2, 3, 5, 8, 13, 21 …
How might you program something to print the Nth term of the Fibonacci sequence?
Iterative Solution: Fibonacci Sequence
An iterative solution using a for loop may look something like this:
1 function getNthFibo(n) {
2 if ( n <= 1) return n;
3 var sum = 0,
4 last = 1,
5 lastlast = 0;
6
7 for (var i = 1; i < n; i++) {
8 sum = lastlast + last;
9 lastlast = last;
10 last = sum;
11 }
12 return sum;
13 }
101
Chapter 8 reCursion
A for loop can be used to keep track of the last two elements of the Fibonacci
sequence, and its sum yields the Fibonacci number.
Now, how might this be done recursively?
Recursive Solution: Fibonacci
The following shows the recursive solution:
1 function getNthFibo(n) {
2 if (n <= 1) {
3 return n;
4 } else {
5 return getNthFibo(n - 1) + getNthFibo(n - 2);
6 }
7 }
Base case: The base case for the Fibonacci sequence is that the first element is 1.
Divide and conquer: By definition of the Fibonacci sequence, the Nth Fibonacci
number is the sum of the (n-1)th and (n-2)th Fibonacci numbers. However, this
implementation has a time complexity of O(2n), which is discussed in detail later in this
chapter. We will explore a more efficient recursive algorithm for the Fibonacci sequence
using tail recursion in the next section.
Fibonacci Sequence: Tail Recursion
A tail recursive function is a recursive function in which the recursive call is the last
executed thing in the function. First let’s look at the iterative solution:
1 function getNthFibo(n) {
2 if ( n <= 1) return n;
3 var sum = 0,
4 last = 1,
5 lastlast = 0;
6
7 for (var i = 1; i < n; i++) {
8 sum = lastlast + last;
9 lastlast = last;
102
Chapter 8 reCursion
10 last = sum;
11 }
12 return sum;
13 }
At each iteration, the following update happens: (lastlast, last) = (last,
lastlast+last). With this structure, the following recursive function can be formed:
1 function getNthFiboBetter(n, lastlast, last) {
2 if (n == 0) {
3 return lastlast;
4 }
5 if (n == 1) {
6 return last;
7 }
8 return getNthFiboBetter(n-1, last, lastlast + last);
9 }
Time Complexity: O(n)
At most, this function executes n times because it’s decremented by n-1 each time
with only single recursive call.
Space Complexity: O(n)
The space complexity is also O(n) because of the stack call used for this function.
This will be further explained in the “Recursive Call Stack Memory” section later in this
chapter.
To conclude the rules of recursion, let’s examine another example, which is more
complex.
Pascal’s Triangle
In this example, a function for calculating a term of Pascal’s triangle will be explored.
Pascal’s triangle is a triangle whose element value is the summation of its top two (left
and right) values, as shown in Figure 8-2.
103
Chapter 8 reCursion
Figure 8-2. Pascal’s triangle
Base case: The base case for Pascal’s triangle is that the top element (row=1, col=1)
is 1. Everything else is derived from this fact alone. Hence, when the column is 1, return 1,
and when the row is 0, return 0.
Divide and conquer: By the mathematical definition of Pascal’s triangle, a term of
Pascal’s triangle is defined as the sum of its upper terms. Therefore, this can be expressed as
the following: pascalTriangle(row - 1, col) + pascalTriangle(row - 1, col - 1).
1 function pascalTriangle(row, col) {
2 if (col == 0) {
3 return 1;
4 } else if (row == 0) {
5 return 0;
6 } else {
7 return pascalTriangle(row - 1, col) + pascalTriangle(row - 1,
col - 1);
8 }
104
Chapter 8 reCursion
9 }
10 pascalTriangle(5, 2); // 10
This is the beauty of recursion! Look next at how short and elegant this code is.
Big-O for Recursion
In Chapter 1, Big-O analysis of recursive algorithms was not covered. This was because
recursive algorithms are much harder to analyze. To perform Big-O analysis for recursive
algorithms, its recurrence relations must be analyzed.
Recurrence Relations
In algorithms implemented iteratively, Big-O analysis is much simpler because loops
clearly define when to stop and how much to increment in each iteration. For analyzing
recursive algorithms, recurrence relations are used. Recurrence relations consist of two-
part analysis: Big-O for base case and Big-O for recursive case.
Let’s revisit the naive Fibonacci sequence example:
function getNthFibo(n) {
if (n <= 1) {
return n;
} else {
return getNthFibo(n - 1) + getNthFibo(n - 2);
}
}
getNthFibo(3);
The base case has a time complexity of O(1). The recursive case calls itself twice.
Let’s represent this as T (n) = T (n − 1) + T (n − 2) + O(1).
• Base case: T (n) = O(1)
• Recursive case: T (n) = T (n − 1) + T (n − 2) + O(1)
Now, this relation means that since T (n) = T (n − 1) + T (n − 2) + O(1), then (by
replacing n with n−1), T (n − 1) = T (n − 2) + T (n − 3) + O(1). Replacing n−1 with n−2
yields T (n − 2) = T (n − 3) + T (n − 4) + O(1). Therefore, you can see that for every call,
there are two more calls for each call. In other words, this has a time complexity of O(2n).
105
Chapter 8 reCursion
It helps to visualize it as such:
F(6) * <-- only once
F(5) *
F(4) **
F(3) ****
F(2) ********
F(1) **************** <-- 16
F(0) ******************************** <-- 32
Calculating Big-O this way is difficult and prone to error. Thankfully, there is a
concept known as the master theorem to help. The master theorem helps programmers
easily analyze the time and space complexities of recursive algorithms.
Master Theorem
The master theorem states the following:
Given a recurrence relation of the form T (n) = aT (n/b) + O(nc)
where a >= 1 and b >=1, there are three cases.
a is the coefficient that is multiplied by the recursive call. b is the “logarithmic” term,
which is the term that divides the n during the recursive call. Finally, c is the polynomial
term on the nonrecursive component of the equation.
The first case is when the polynomial term on the nonrecursive component O(nc) is
smaller than logb(a).
Case 1: c < logb(a) then T (n) = O(n(logb(a))).
For example, T (n) = 8T (n/2) + 1000n2
Identify a, b, c: a = 8, b = 2, c = 2
Evaluate: log2(8) = 3. c < 3 is satisfied.
Result: T (n) = O(n3)
The second case is when c is equal to logb(a).
Case 2: c = logb(a) then T (n) = O(nclog(n)).
For example, T (n) = 2T (n/2) + 10n.
106
Chapter 8 reCursion
Identify a, b, c: a = 2, b = 2, c = 1
Evaluate: log2(2) = 1. c = 1 is satisfied.
Result: T (n) = O(nclog(n)) = T (n) = O(n1log(n)) = T (n) = O(nlog(n))
The third and final case is when c is greater than logb(a).
Case 3: c > logb(a) then T (n) = O(f (n)).
For example, T (n) = 2T (n/2) + n2.
Identify a,b,c: a = 2, b = 2, c = 2
Evaluate: log2(2) = 1. c > 1 is satisfied.
Result: T (n) = f (n) = O(n2)
This section covered a lot about analyzing the time complexity of recursive
algorithms. Space complexity analysis is just as important. The memory used by
recursive function calls should also be noted and analyzed for space complexity
analysis.
Recursive Call Stack Memory
When a recursive function calls itself, that takes up memory, and this is really important
in Big-O space complexity analysis.
For example, this simple function for printing from n to 1 recursively takes O(n) in
space:
1 function printNRecursive(n) {
2 [Link](n);
3 if (n > 1){
4 printNRecursive(n-1);
5 }
6 }
7 printNRecursive(10);
A developer can run this on a browser or any JavaScript engine and will see the result
shown in Figure 8-3 in the call stack.
107
Chapter 8 reCursion
Figure 8-3. Call stack in Developer Tools
As shown in Figures 8-3 and 8-4, each recursive call must be stored in memory until
the base case is resolved. Recursive algorithms take extra memory because of the call
stack.
Figure 8-4. Call stack memory
Recursive functions have an additional space complexity cost that comes from the
recursive calls that need to be stored in the operating system’s memory stack. The stack
is accumulated until the base case is solved. In fact, this is often why an iterative solution
108
Chapter 8 reCursion
may be preferred over the recursive solution. In the worst case, if the base case is
implemented incorrectly, the recursive function will cause the program to crash because
of a stack overflow error that occurs when there are more than the allowed number of
elements in the memory stack.
Summary
Recursion is a powerful tool to implement complex algorithms. Recall that all recursive
functions consist of two parts: the base case and the divide-and-conquer method
(solving subproblems).
Analyzing the Big-O of these recursive algorithms can be done empirically (not
recommended) or by using the master theorem. Recall that the master theorem needs
the recurrence relation in the following form: T (n) = aT (n/b) +O(nc). When using the
master theorem, identify a, b, and c to determine which of the three cases of the master
theorem it belongs to.
Finally, when implementing and analyzing recursive algorithms, consider the
additional memory caused by the call stack of the recursive function calls. Each recursive
call requires a place in the call stack at runtime; when the call stack accumulate n calls,
then the space complexity of the function is O(n).
Exercises
These exercises on recursion cover varying problems to help solidify the knowledge
gained from this chapter. The focus should be to identify the correct base case first before
solving the entire problem. You will find all the code for the exercises on GitHub.1
CONVERT DECIMAL (BASE 10) TO BINARY NUMBER
to do this, keep dividing the number by 2 and each time calculate the modulus (remainder)
and division.
Base case: the base case for this problem is when the n is less than 2. When it is less
than 2, it can be only 0 or 1.
109
Chapter 8 reCursion
1 function base10ToString(n) {
2 var binaryString = "";
3
4 function base10ToStringHelper(n) {
5 if (n < 2) {
6 binaryString += n;
7 return;
8 } else {
9 base10ToStringHelper([Link](n / 2));
10 base10ToStringHelper(n % 2);
11 }
12 }
13 base10ToStringHelper(n);
14
15 return binaryString;
16 }
17
18 [Link](base10ToString(232)); // 11101000
Time Complexity: O(log2(n))
time complexity is logarithmic because the recursive call divides the n by 2, which makes the
algorithm fast. For example, for n = 8, it executes only three times. For n=1024, it executes
10 times.
Space Complexity: O(log2(n))
PRINT ALL PERMUTATIONS OF AN ARRAY
this is a classical recursion problem and one that is pretty hard to solve. the premise of the
problem is to swap elements of the array in every possible position.
First, let’s draw the recursion tree for this problem (see Figure 8-5).
110
Chapter 8 reCursion
Figure 8-5. Permutation of array recursion tree
Base case: beginIndex is equal to endIndex.
When this occurs, the function should print the current permutation.
Permutations: We will need a function to swap elements:
1 function swap(strArr, index1, index2) {
2 var temp = strArr[index1];
3 strArr[index1] = strArr[index2];
4 strArr[index2] = temp;
5 }
1 function permute(strArr, begin, end) {
2 if (begin == end) {
3 [Link](strArr);
4 } else {
5 for (var i = begin; i < end + 1; i++) {
6 swap(strArr, begin, i);
7 permute(strArr, begin + 1, end);
8 swap(strArr, begin, i);
9 }
111
Chapter 8 reCursion
10 }
11 }
12
13 function permuteArray(strArr) {
14 permute(strArr, 0, [Link] - 1);
15 }
16
17 permuteArray(["A", "C", "D"]);
18 // ["A", "C", "D"]
19 // ["A", "D", "C"]
20 // ["C", "A", "D"]
21 // ["C", "D", "A"]
22 // ["D", "C", "A"]
23 // ["D", "A", "C"]
Time Complexity: O(n!)
Space Complexity: O(n!)
there are n! permutations, and it creates n! call stacks.
FLATTEN AN OBJECT
Given a Javascript array like this:
1 var dictionary = {
2 'Key1': '1',
3 'Key2': {
4 'a' : '2',
5 'b' : '3',
6 'c' : {
7 'd' : '3',
8 'e' : '1'
9 }
10 }
11 }
112
Chapter 8 reCursion
flatten it into {'Key1': '1', 'Key2.a': '2','Key2.b' : '3', 'Key2.c.d' : '3',
'Key2.c.e' : '1'}, where the child is denoted by . between the parent and child (see
Figure 8-6).
to do this, iterate over any property and recursively check it for child properties, passing in the
concatenated string name.
Figure 8-6. Flatten a dictionary recursion tree
Base case: the base case for this problem is when input is not an object.
1 function flattenDictionary(dictionary) {
2 var flattenedDictionary = {};
3
4 function flattenDitionaryHelper(dictionary, propName) {
5 if (typeof dictionary != 'object') {
6 flattenedDictionary[propName] = dictionary;
7 return;
8 }
9 for (var prop in dictionary) {
10 if (propName == "){
11 flattenDitionaryHelper(dictionary[prop], propName+prop);
12 } else {
113
Chapter 8 reCursion
13 flattenDitionaryHelper(dictionary[prop],
propName+'.'+prop);
14 }
15 }
16 }
17
18 flattenDitionaryHelper(dictionary, ");
19 return flattenedDictionary;
20 }
Time Complexity: O(n)
Space Complexity: O(n)
each property is visited only once and stored once per n properties.
WRITE A PROGRAM THAT RECURSIVELY DETERMINES IF A STRING IS A PALINDROME
a palindrome is a word spelled the same backward and forward such as deified,
racecar, testset, and aibohphobia (the fear of palindromes).
1 function isPalindromeRecursive(word) {
2 return isPalindromeHelper(word, 0, [Link]-1);
3 }
4
5 function isPalindromeHelper(word, beginPos, endPos) {
6 if (beginPos >= endPos) {
7 return true;
8 }
9 if ([Link](beginPos) != [Link](endPos)) {
10 return false;
11 } else {
12 return isPalindromeHelper(word, beginPos + 1, endPos - 1);
13 }
14 }
15
16 isPalindromeRecursive('hi'); // false
17 isPalindromeRecursive('iii'); // true
114
Chapter 8 reCursion
18 isPalindromeRecursive('ii'); // true
19 isPalindromeRecursive('aibohphobia'); // true
20 isPalindromeRecursive('racecar'); // true
the idea behind this one is that with two indexes (one in front and one in back) you check at
each step until the front and back meet.
Time Complexity: O(n)
Space Complexity: O(n)
space complexity here is still o(n) because of the recursive call stack. remember that the call
stack remains part of memory even if it is not declaring a variable or being stored inside a data
structure.
115