1x

A Spanning Tree (ST) of a connected undirected weighted graph G is a subgraph of G that is a tree and connects (spans) all vertices of G. A graph G can have many STs (see this or this), each with different total weight (the sum of edge weights in the ST).

A Min(imum) Spanning Tree (MST) of G is an ST of G that has the smallest total weight among the various STs.

Remarks: By default, we show e-Lecture Mode for first time (or non logged-in) visitor.
If you are an NUS student and a repeat visitor, please login.

🕑

The MST problem is a standard graph (and also optimization) problem defined as follows: Given a connected undirected weighted graph G = (V, E), select a subset of edges of G such that the graph is still connected but with minimum total weight. The output is either the actual MST of G (there can be several possible MSTs of G) or usually just the minimum total weight itself (this is unique).

Pro-tip 1: Since you are not logged-in, you may be a first time visitor (or not an NUS student) who are not aware of the following keyboard shortcuts to navigate this e-Lecture mode: [PageDown]/[PageUp] to go to the next/previous slide, respectively, (and if the drop-down box is highlighted, you can also use [→ or ↓/← or ↑] to do the same),and [Esc] to toggle between this e-Lecture mode and exploration mode.

🕑

Government wants to link N rural villages in the country with N-1 roads.
(that is a spanning tree with N vertices and N-1 edges).

The cost to build a road to connect two villages depends on the terrain, distance, etc.
(that is a complete undirected weighted graph of N*(N-1)/2 weighted edges).

You want to minimize the total building cost. How are you going to build the roads?
(that is minimum spanning tree).

PS: There is a variant of this problem that requires more advanced solution, e.g., see this.

Pro-tip 2: We designed this visualization and this e-Lecture mode to look good on 1366x768 resolution or larger (typical modern laptop resolution in 2021). We recommend using Google Chrome to access VisuAlgo. Go to full screen mode (F11) to enjoy this setup. However, you can use zoom-in (Ctrl +) or zoom-out (Ctrl -) to calibrate this.

🕑

The MST problem has polynomial solutions.

In this visualization, we will learn two of them: Kruskal's algorithm and Prim's algorithm. Both are classified as Greedy Algorithms. Note that there are other MST algorithms outside the two presented here.

Pro-tip 3: Other than using the typical media UI at the bottom of the page, you can also control the animation playback using keyboard shortcuts (in Exploration Mode): Spacebar to play/pause/replay the animation, / to step the animation backwards/forwards, respectively, and -/+ to decrease/increase the animation speed, respectively.

🕑

View the visualisation of MST algorithm above.

Originally, all vertices and edges in the input graph are colored with the standard black color on white background.

At the end of the MST algorithm, |V|-1 MST edges (and all |V| vertices) will be colored orange and non-MST edges will be colored grey.

🕑

There are two different sources for specifying an input graph:

1. Edit Graph: You can edit the currently displayed connected undirected weighted graph or draw your own input graph.
2. Example Graphs: You can select from the list of example connected undirected weighted graphs to get you started.
🕑

Kruskal's algorithm: An O(E log V) greedy MST algorithm that grows a forest of minimum spanning trees and eventually combine them into one MST.

Kruskal's requires a good sorting algorithm to sort edges of the input graph (usually stored in an Edge List data structure) by non-decreasing weight and another data structure called Union-Find Disjoint Sets (UFDS) to help in checking/preventing cycle.

🕑

Kruskal's algorithm first sort the set of edges E in non-decreasing weight (there can be edges with the same weight), and if ties, by increasing smaller vertex number of the edge, and if still ties, by increasing larger vertex number of the edge.

Discussion: Is this the only possible sort criteria?

🕑

The content of this interesting slide (the answer of the usually intriguing discussion point from the earlier slide) is hidden and only available for legitimate CS lecturer worldwide. This mechanism is used in the various flipped classrooms in NUS.

If you are really a CS lecturer (or an IT teacher) (outside of NUS) and are interested to know the answers, please drop an email to stevenhalim at gmail dot com (show your University staff profile/relevant proof to Steven) for Steven to manually activate this CS lecturer-only feature for you.

FAQ: This feature will NOT be given to anyone else who is not a CS lecturer.

🕑

Then, Kruskal's algorithm will perform a loop through these sorted edges (that already have non-decreasing weight property) and greedily taking the next edge e if it does not create any cycle w.r.t. edges that have been taken earlier.

Without further ado, let's try Kruskal on the default example graph (that has three edges with the same weight). Go through this animated example first before continuing.

🕑

To see on why the Greedy Strategy of Kruskal's algorithm works, we define a loop invariant: Every edge e that is added into tree T by Kruskal's algorithm is part of the MST.

At the start of Kruskal's main loop, T = {} is always part of MST by definition.

Kruskal's has a special cycle check in its main loop (using UFDS data structure) and only add an edge e into T if it will never form a cycle w.r.t. the previously selected edges.

At the end of the main loop, Kruskal's can only select V-1 edges from a connected undirected weighted graph G without having any cycle. This implies that Kruskal's produces a Spanning Tree.

On the default example, notice that after taking the first 2 edges: 0-1 and 0-3, in that order, Kruskal's cannot take edge 1-3 as it will cause a cycle 0-1-3-0. Kruskal's then take edge 0-2 but it cannot take edge 2-3 as it will cause cycle 0-2-3-0.

🕑

We have seen in the previous slide that Kruskal's algorithm will produce a tree T that is a Spanning Tree (ST) when it stops. But is it the minimum ST, i.e., the MST?

To prove this, we need to recall that before running Kruskal's main loop, we have already sort the edges in non-decreasing weight, i.e., the latter edges will have equal or larger weight than the earlier edges.

🕑

At the start of every loop, T is always part of MST.

If Kruskal's only add a legal edge e (that will not cause cycle w.r.t. the edges that have been taken earlier) with min cost, then we can be sure that w(T U e) ≤ w(T U any other unprocessed edge e' that does not form cycle) (by virtue that Kruskal's has sorted the edges, so w(e) ≤ w(e')).

Therefore, at the end of the loop, the Spanning Tree T must have minimal overall weight w(T), so T is the final MST.

On the default example, notice that after taking the first 2 edges: 0-1 and 0-3, in that order, and ignoring edge 1-3 as it will cause a cycle 0-1-3-0, we can safely take the next smallest legal edge 0-2 (with weight 2) as taking any other legal edge (e.g., edge 2-3 with larger weight 3) will either create another MST with equal weight (not in this example) or another ST that is not minimum (which is this example).

🕑

There are two parts of Kruskal's algorithm: Sorting and the Kruskal's main loop.

The sorting of edges is easy. We just store the graph using Edge List data structure and sort E edges using any O(E log E) = O(E log V) sorting algorithm (or just use C++/Python/Java sorting library routine) by non-decreasing weight, smaller vertex number, higher vertex number. This O(E log V) is the bottleneck part of Kruskal's algorithm as the second part is actually lighter, see below.

Kruskal's main loop can be easily implemented using Union-Find Disjoint Sets data structure. We use IsSameSet(u, v) to test if taking edge e with endpoints u and v will cause a cycle (same connected component -- there is another path in the subtree that can connect u to v, thus adding edge (u, v) will cause a cycle) or not. If IsSameSet(u, v) returns false, we greedily take this next smallest and legal edge e and call UnionSet(u, v) to prevent future cycles involving this edge. This part runs in O(E) as we assume UFDS IsSameSet(u, v) and UnionSet(u, v) operations run in O(1) for a relatively small graph.

🕑

Prim's algorithm: Another O(E log V) greedy MST algorithm that grows a Minimum Spanning Tree from a starting source vertex until it spans the entire graph.

Prim's requires a Priority Queue data structure (usually implemented using Binary Heap but we can also use Balanced Binary Search Tree too) to dynamically order the currently considered edges based on non-decreasing weight, an Adjacency List data structure for fast neighbor enumeration of a vertex, and a Boolean array (a Direct Addressing Table) to help in checking cycle.

Another name of Prim's algorithm is Jarnik-Prim's algorithm.

🕑

Prim's algorithm starts from a designated source vertex s (usually vertex 0) and enqueues all edges incident to s into a Priority Queue (PQ) according to non-decreasing weight, and if ties, by increasing vertex number (of the neighboring vertex number). Then it will repeatedly do the following greedy steps: If the vertex v of the front-most edge pair information e: (w, v) in the PQ has not been visited, it means that we can greedily extends the tree T to include vertex v and enqueue edges connected to v into the PQ, otherwise we discard edge e (because Prim's grows one spanning tree from s, the fact that v is already visited implies that there is another path from s to v and adding this edge will cause a cycle).

Without further ado, let's try Prim(1) on the default example graph (that has three edges with the same weight). That's it, we start Prim's algorithm from source vertex s = 1. Go through this animated example first before continuing.

🕑

Prim's algorithm is a Greedy Algorithm because at each step of its main loop, it always try to select the next valid edge e with minimal weight (that is greedy!).

To convince us that Prim's algorithm is correct, let's go through the following simple proof: Let T be the spanning tree of graph G generated by Prim's algorithm and T* be the spanning tree of G that is known to have minimal cost, i.e. T* is the MST.

If T == T*, that's it, Prim's algorithm produces exactly the same MST as T*, we are done.

But if T != T*...

🕑

Assume that on the default example, T = {0-1, 0-3, 0-2} but T* = {0-1, 1-3, 0-2} instead.

Let ek = (u, v) be the first edge chosen by Prim's Algorithm at the k-th iteration that is not in T* (on the default example, k = 2, e2 = (0, 3), note that (0, 3) is not in T*).

Let P be the path from u to v in T*, and let e* be an edge in P such that one endpoint is in the tree generated at the (k−1)-th iteration of Prim's algorithm and the other is not (on the default example, P = 0-1-3 and e* = (1, 3), note that vertex 1 is inside T at first iteration k = 1).

🕑

If the weight of e* is less than the weight of ek, then Prim's algorithm would have chosen e* on its k-th iteration as that is how Prim's algorithm works.

So, it is certain that w(e*) ≥ w(ek).
(on the example graph, e* = (1, 3) has weight 1 and ek = (0, 3) also has weight 1).

When weight e* is = weight ek, the choice between the e* or ek is actually arbitrary. And whether the weight of e* is ≥ weight of ek, e* can always be substituted with ek while preserving minimal total weight of T*. (on the example graph, when we replace e* = (1, 3) with ek = (0, 3), we manage to transform T* into T).

🕑

But if T != T*... (continued)

We can repeat the substitution process outlined earlier repeatedly until T* = T and thereby we have shown that the spanning tree generated by any instance of Prim's algorithm (from any source vertex s) is an MST as whatever the optimal MST is, it can be transformed to the output of Prim's algorithm.

🕑

We can easily implement Prim's algorithm with two well-known data structures:

1. A Priority Queue PQ (Binary Heap inside C++ STL priority_queue/Python heapq/Java PriorityQueue or Balanced BST inside C++ STL set/Java TreeSet), and
2. A Boolean array of size V, essentially a Direct Addressing Table (to decide if a vertex has been taken or not, i.e., in the same connected component as the source vertex s or not).

With these, we can run Prim's Algorithm in O(E log V) because we process each edge once and each time, we call Insert((w, v)) and (w, v) = ExtractMax() from a PQ in O(log E) = O(log V2) = O(2 log V) = O(log V). As there are E edges, Prim's Algorithm runs in O(E log V).

🕑

Quiz: Having seen both Kruskal's and Prim's Algorithms, which one is the better MST algorithm?

Kruskal's Algorithm
It Depends
Prim's Algorithm

Discussion: Why?

🕑

The content of this interesting slide (the answer of the usually intriguing discussion point from the earlier slide) is hidden and only available for legitimate CS lecturer worldwide. This mechanism is used in the various flipped classrooms in NUS.

If you are really a CS lecturer (or an IT teacher) (outside of NUS) and are interested to know the answers, please drop an email to stevenhalim at gmail dot com (show your University staff profile/relevant proof to Steven) for Steven to manually activate this CS lecturer-only feature for you.

FAQ: This feature will NOT be given to anyone else who is not a CS lecturer.

🕑

You have reached the end of the basic stuffs of this Min(imum) Spanning Tree graph problem and its two classic algorithms: Kruskal's and Prim's (there are others, like another O(E log V) Boruvka's algorithm, but not discussed in this visualization). We encourage you to explore further in the Exploration Mode.

However, the harder MST problems can be (much) more challenging that its basic version.

Once you have (roughly) mastered this MST topic, we encourage you to study more on harder graph problems where MST is used as a component, e.g., approximation algorithm for NP-hard (Metric No-Repeat) TSP and Steiner Tree problems.

🕑

We write a few MST problem variants in the Competitive Programming book.

1. Max(imum) Spanning Tree,
2. Min(imum) Spanning Subgraph,
3. Min(imum) Spanning Forest,
4. Second Best Spanning Tree,
5. Minimax (Maximin) Path Problem, etc

Advertisement: Buy CP book to study more about these variants and see that sometimes Kruskal's is better and sometimes Prim's is better at some of these variants.

🕑

For a few more challenging questions about this MST problem and/or Kruskal's/Prim's Algorithms, please practice on MST training module (no login is required, but on medium difficulty setting only).

However, for NUS students, you should login to officially clear this module and such achievement will be recorded in your user account.

🕑

This MST problem can be much more challenging than this basic form. Therefore we encourage you to try the following two ACM ICPC contest problems about MST: UVa 01234 - RACING and Kattis - arcticnetwork.

You are allowed to use/modify our implementation code for Kruskal's/Prim's Algorithms:
kruskal.cpp | py | java | ml
prim.cpp | py | java | ml

🕑

The content of this interesting slide (the answer of the usually intriguing discussion point from the earlier slide) is hidden and only available for legitimate CS lecturer worldwide. This mechanism is used in the various flipped classrooms in NUS.

If you are really a CS lecturer (or an IT teacher) (outside of NUS) and are interested to know the answers, please drop an email to stevenhalim at gmail dot com (show your University staff profile/relevant proof to Steven) for Steven to manually activate this CS lecturer-only feature for you.

FAQ: This feature will NOT be given to anyone else who is not a CS lecturer.

You have reached the last slide. Return to 'Exploration Mode' to start exploring!

Note that if you notice any bug in this visualization or if you want to request for a new visualization feature, do not hesitate to drop an email to the project leader: Dr Steven Halim via his email address: stevenhalim at gmail dot com.

🕑

Visualisation Scale

Toggle V. Number for 0.5x

Edit Graph

Example Graphs

Kruskal's Algorithm

Prim's Algorithm(s)

>

1.0x (Default)

0.5x (Minimal Details)

CP 4.10

CP 4.14

K5

Rail

Tessellation

Large

s =

Go

Initially conceived in 2011 by Associate Professor Steven Halim, VisuAlgo aimed to facilitate a deeper understanding of data structures and algorithms for his students by providing a self-paced, interactive learning platform.

Featuring numerous advanced algorithms discussed in Dr. Steven Halim's book, 'Competitive Programming' — co-authored with Dr. Felix Halim and Dr. Suhendry Effendy — VisuAlgo remains the exclusive platform for visualizing and animating several of these complex algorithms even after a decade.

While primarily designed for National University of Singapore (NUS) students enrolled in various data structure and algorithm courses (e.g., CS1010/equivalent, CS2040/equivalent (including IT5003), CS3230, CS3233, and CS4234), VisuAlgo also serves as a valuable resource for inquisitive minds worldwide, promoting online learning.

Initially, VisuAlgo was not designed for small touch screens like smartphones, as intricate algorithm visualizations required substantial pixel space and click-and-drag interactions. For an optimal user experience, a minimum screen resolution of 1366x768 is recommended. However, since April 2022, a mobile (lite) version of VisuAlgo has been made available, making it possible to use a subset of VisuAlgo features on smartphone screens.

VisuAlgo remains a work in progress, with the ongoing development of more complex visualizations. At present, the platform features 24 visualization modules.

Equipped with a built-in question generator and answer verifier, VisuAlgo's "online quiz system" enables students to test their knowledge of basic data structures and algorithms. Questions are randomly generated based on specific rules, and students' answers are automatically graded upon submission to our grading server. As more CS instructors adopt this online quiz system worldwide, it could effectively eliminate manual basic data structure and algorithm questions from standard Computer Science exams in many universities. By assigning a small (but non-zero) weight to passing the online quiz, CS instructors can significantly enhance their students' mastery of these basic concepts, as they have access to an almost unlimited number of practice questions that can be instantly verified before taking the online quiz. Each VisuAlgo visualization module now includes its own online quiz component.

VisuAlgo has been translated into three primary languages: English, Chinese, and Indonesian. Additionally, we have authored public notes about VisuAlgo in various languages, including Indonesian, Korean, Vietnamese, and Thai:

id, kr, vn, th.

Team

Associate Professor Steven Halim, School of Computing (SoC), National University of Singapore (NUS)
Dr Felix Halim, Senior Software Engineer, Google (Mountain View)

CDTL TEG 1: Jul 2011-Apr 2012: Koh Zi Chun, Victor Loh Bo Huai

Final Year Project/UROP students 1
Jul 2012-Dec 2013: Phan Thi Quynh Trang, Peter Phandi, Albert Millardo Tjindradinata, Nguyen Hoang Duy
Jun 2013-Apr 2014 Rose Marie Tan Zhao Yun, Ivan Reinaldo

CDTL TEG 2: May 2014-Jul 2014: Jonathan Irvin Gunawan, Nathan Azaria, Ian Leow Tze Wei, Nguyen Viet Dung, Nguyen Khac Tung, Steven Kester Yuwono, Cao Shengze, Mohan Jishnu

Final Year Project/UROP students 2
Jun 2014-Apr 2015: Erin Teo Yi Ling, Wang Zi
Jun 2016-Dec 2017: Truong Ngoc Khanh, John Kevin Tjahjadi, Gabriella Michelle, Muhammad Rais Fathin Mudzakir
Aug 2021-Apr 2023: Liu Guangyuan, Manas Vegi, Sha Long, Vuong Hoang Long, Ting Xiao, Lim Dewen Aloysius

Optiver: Aug 2023-Oct 2023: Bui Hong Duc, Oleh Naver, Tay Ngan Lin

Final Year Project/UROP students 3
Aug 2023-Apr 2024: Xiong Jingya, Radian Krisno, Ng Wee Han

List of translators who have contributed ≥ 100 translations can be found at statistics page.

Acknowledgements
NUS CDTL gave Teaching Enhancement Grant to kickstart this project.

For Academic Year 2023/24, a generous donation from Optiver will be used to further develop VisuAlgo.

VisuAlgo is generously offered at no cost to the global Computer Science community. If you appreciate VisuAlgo, we kindly request that you spread the word about its existence to fellow Computer Science students and instructors. You can share VisuAlgo through social media platforms (e.g., Facebook, YouTube, Instagram, TikTok, Twitter, etc), course webpages, blog reviews, emails, and more.

Data Structures and Algorithms (DSA) students and instructors are welcome to use this website directly for their classes. If you capture screenshots or videos from this site, feel free to use them elsewhere, provided that you cite the URL of this website (https://visualgo.net) and/or the list of publications below as references. However, please refrain from downloading VisuAlgo's client-side files and hosting them on your website, as this constitutes plagiarism. At this time, we do not permit others to fork this project or create VisuAlgo variants. Personal use of an offline copy of the client-side VisuAlgo is acceptable.

Please note that VisuAlgo's online quiz component has a substantial server-side element, and it is not easy to save server-side scripts and databases locally. Currently, the general public can access the online quiz system only through the 'training mode.' The 'test mode' offers a more controlled environment for using randomly generated questions and automatic verification in real examinations at NUS.

List of Publications

This work has been presented at the CLI Workshop at the ICPC World Finals 2012 (Poland, Warsaw) and at the IOI Conference at IOI 2012 (Sirmione-Montichiari, Italy). You can click this link to read our 2012 paper about this system (it was not yet called VisuAlgo back in 2012) and this link for the short update in 2015 (to link VisuAlgo name with the previous project).

Bug Reports or Request for New Features

VisuAlgo is not a finished project. Associate Professor Steven Halim is still actively improving VisuAlgo. If you are using VisuAlgo and spot a bug in any of our visualization page/online quiz tool or if you want to request for new features, please contact Associate Professor Steven Halim. His contact is the concatenation of his name and add gmail dot com.