Linear Programming and the Simplex Algorithm

In the last post in this series we saw some simple examples of linear programs, derived the concept of a dual linear program, and saw the duality theorem and the complementary slackness conditions which give a rough sketch of the stopping criterion for an algorithm. This time we’ll go ahead and write this algorithm for solving linear programs, and next time we’ll apply the algorithm to an industry-strength version of the nutrition problem we saw last time. The algorithm we’ll implement is called the simplex algorithm. It was the first algorithm for solving linear programs, invented in the 1940’s by George Dantzig, and it’s still the leading practical algorithm, and it was a key part of a Nobel Prize. It’s by far one of the most important algorithms ever devised.

As usual, we’ll post all of the code written in the making of this post on this blog’s Github page.

Slack variables and equality constraints

The simplex algorithm can solve any kind of linear program, but it only accepts a special form of the program as input. So first we have to do some manipulations. Recall that the primal form of a linear program was the following minimization problem.

\min \left \langle c, x \right \rangle \\ \textup{s.t. } Ax \geq b, x \geq 0

where the brackets mean “dot product.” And its dual is

\max \left \langle y, b \right \rangle \\ \textup{s.t. } A^Ty \leq c, y \geq 0

The linear program can actually have more complicated constraints than just the ones above. In general, one might want to have “greater than” and “less than” constraints in the same problem. It turns out that this isn’t any harder, and moreover the simplex algorithm only uses equality constraints, and with some finicky algebra we can turn any set of inequality or equality constraints into a set of equality constraints.

We’ll call our goal the “standard form,” which is as follows:

\max \left \langle c, x \right \rangle \\ \textup{s.t. } Ax = b, x \geq 0

It seems impossible to get the usual minimization/maximization problem into standard form until you realize there’s nothing stopping you from adding more variables to the problem. That is, say we’re given a constraint like:

\displaystyle x_7 + x_3 \leq 10,

we can add a new variable \xi, called a slack variable, so that we get an equality:

\displaystyle x_7 + x_3 + \xi = 10

And now we can just impose that \xi \geq 0. The idea is that \xi represents how much “slack” there is in the inequality, and you can always choose it to make the condition an equality. So if the equality holds and the variables are nonnegative, then the x_i will still satisfy their original inequality. For “greater than” constraints, we can do the same thing but subtract a nonnegative variable. Finally, if we have a minimization problem “\min z” we can convert it to \max -z.

So, to combine all of this together, if we have the following linear program with each kind of constraint,

Screen Shot 2014-10-05 at 12.06.19 AM

We can add new variables \xi_1, \xi_2, and write it as

Screen Shot 2014-10-05 at 12.06.41 AM

By defining the vector variable x = (x_1, x_2, x_3, \xi_1, \xi_2) and c = (-1,-1,-1,0,0) and A to have -1, 0, 1 as appropriately for the new variables, we see that the system is written in standard form.

This is the kind of tedious transformation we can automate with a program. Assuming there are n variables, the input consists of the vector c of length n, and three matrix-vector pairs (A, b) representing the three kinds of constraints. It’s a bit annoying to describe, but the essential idea is that we compute a rectangular “identity” matrix whose diagonal entries are \pm 1, and then join this with the original constraint matrix row-wise. The reader can see the full implementation in the Github repository for this post, though we won’t use this particular functionality in the algorithm that follows.

There are some other additional things we could do: for example there might be some variables that are completely unrestricted. What you do in this case is take an unrestricted variable z and replace it by the difference of two unrestricted variables z' - z''.  For simplicity we’ll ignore this, but it would be a fruitful exercise for the reader to augment the function to account for these.

What happened to the slackness conditions?

The “standard form” of our linear program raises an obvious question: how can the complementary slackness conditions make sense if everything is an equality? It turns out that one can redo all the work one did for linear programs of the form we gave last time (minimize w.r.t. greater-than constraints) for programs in the new “standard form” above. We even get the same complementary slackness conditions! If you want to, you can do this entire routine quite a bit faster if you invoke the power of Lagrangians. We won’t do that here, but the tool shows up as a way to work with primal-dual conversions in many other parts of mathematics, so it’s a good buzzword to keep in mind.

In our case, the only difference with the complementary slackness conditions is that one of the two is trivial: \left \langle y^*, Ax^* - b \right \rangle = 0. This is because if our candidate solution x^* is feasible, then it will have to satisfy Ax = b already. The other one, that \left \langle x^*, A^Ty^* - c \right \rangle = 0, is the only one we need to worry about.

Again, the complementary slackness conditions give us inspiration here. Recall that, informally, they say that when a variable is used at all, it is used as much as it can be to fulfill its constraint (the corresponding dual constraint is tight). So a solution will correspond to a choice of some variables which are either used or not, and a choice of nonzero variables will correspond to a solution. We even saw this happen in the last post when we observed that broccoli trumps oranges. If we can get a good handle on how to navigate the set of these solutions, then we’ll have a nifty algorithm.

Let’s make this official and lay out our assumptions.

Extreme points and basic solutions

Remember that the graphical way to solve a linear program is to look at the line (or hyperplane) given by \langle c, x \rangle = q and keep increasing q (or decreasing it, if you are minimizing) until the very last moment when this line touches the region of feasible solutions. Also recall that the “feasible region” is just the set of all solutions to Ax = b, that is the solutions that satisfy the constraints. We imagined this picture:

The constraints define a convex area of "feasible solutions." Image source: Wikipedia.

The constraints define a convex area of “feasible solutions.” Image source: Wikipedia.

With this geometric intuition it’s clear that there will always be an optimal solution on a vertex of the feasible region. These points are called extreme points of the feasible region. But because we will almost never work in the plane again (even introducing slack variables makes us relatively high dimensional!) we want an algebraic characterization of these extreme points.

If you have a little bit of practice with convex sets the correct definition is very natural. Recall that a set X is convex if for any two points x, y \in X every point on the line segment between x and y is also in X. An algebraic way to say this (thinking of these points now as vectors) is that every point \delta x + (1-\delta) y \in X when 0 \leq \delta \leq 1. Now an extreme point is just a point that isn’t on the inside of any such line, i.e. can’t be written this way for 0 < \delta < 1. For example,

A convex set with extremal points in red. Image credit Wikipedia.

A convex set with extremal points in red. Image credit Wikipedia.

Another way to say this is that if z is an extreme point then whenever z can be written as \delta x + (1-\delta) y for some 0 < \delta < 1, then actually x=y=z. Now since our constraints are all linear (and there are a finite number of them) they won’t define a convex set with weird curves like the one above. This means that there are a finite number of extreme points that just correspond to the intersections of some of the constraints. So there are at most 2^n possibilities.

Indeed we want a characterization of extreme points that’s specific to linear programs in standard form, “\max \langle c, x \rangle \textup{ s.t. } Ax=b, x \geq 0.” And here is one.

Definition: Let A be an m \times n matrix with n \geq m. A solution x to Ax=b is called basic if at most m of its entries are nonzero.

The reason we call it “basic” is because, under some mild assumptions we describe below, a basic solution corresponds to a vector space basis of \mathbb{R}^m. Which basis? The one given by the m columns of A used in the basic solution. We don’t need to talk about bases like this, though, so in the event of a headache just think of the basis as a set B \subset \{ 1, 2, \dots, n \} of size m corresponding to the nonzero entries of the basic solution.

Indeed, what we’re doing here is looking at the matrix A_B formed by taking the columns of A whose indices are in B, and the vector x_B in the same way, and looking at the equation A_Bx_B = b. If all the parts of x that we removed were zero then this will hold if and only if Ax=b. One might worry that A_B is not invertible, so we’ll go ahead and assume it is. In fact, we’ll assume that every set of m columns of A forms a basis and that the rows of A are also linearly independent. This isn’t without loss of generality because if some rows or columns are not linearly independent, we can remove the offending constraints and variables without changing the set of solutions (this is why it’s so nice to work with the standard form).

Moreover, we’ll assume that every basic solution has exactly m nonzero variables. A basic solution which doesn’t satisfy this assumption is called degenerate, and they’ll essentially be special corner cases in the simplex algorithm. Finally, we call a basic solution feasible if (in addition to satisfying Ax=b) it satisfies x \geq 0. Now that we’ve made all these assumptions it’s easy to see that choosing m nonzero variables uniquely determines a basic feasible solution. Again calling the sub-matrix A_B for a basis B, it’s just x_B = A_B^{-1}b. Now to finish our characterization, we just have to show that under the same assumptions basic feasible solutions are exactly the extremal points of the feasible region.

Proposition: A vector x is a basic feasible solution if and only if it’s an extreme point of the set \{ x : Ax = b, x \geq 0 \}.

Proof. For one direction, suppose you have a basic feasible solution x, and say we write it as x = \delta y + (1-\delta) z for some 0 < \delta < 1. We want to show that this implies y = z. Since all of these points are in the feasible region, all of their coordinates are nonnegative. So whenever a coordinate x_i = 0 it must be that both y_i = z_i = 0. Since x has exactly n-m zero entries, it must be that y, z both have at least n-m zero entries, and hence y,z are both basic. By our non-degeneracy assumption they both then have exactly m nonzero entries. Let B be the set of the nonzero indices of x. Because Ay = Az = b, we have A(y-z) = 0. Now y-z has all of its nonzero entries in B, and because the columns of A_B are linearly independent, the fact that A_B(y-z) = 0 implies y-z = 0.

In the other direction, suppose  that you have some extreme point x which is feasible but not basic. In other words, there are more than m nonzero entries of x, and we’ll call the indices J = \{ j_1, \dots, j_t \} where t > m. The columns of A_J are linearly dependent (since they’re t vectors in \mathbb{R}^m), and so let \sum_{i=1}^t z_{j_i} A_{j_i} be a nontrivial linear combination of the columns of A. Add zeros to make the z_{j_i} into a length n vector z, so that Az = 0. Now

A(x + \varepsilon z) = A(x - \varepsilon z) = Ax = b

And if we pick \varepsilon sufficiently small x \pm \varepsilon z will still be nonnegative, because the only entries we’re changing of x are the strictly positive ones. Then x = \delta (x + \varepsilon z) + (1 - \delta) \varepsilon z for \delta = 1/2, but this is very embarrassing for x who was supposed to be an extreme point. \square

Now that we know extreme points are the same as basic feasible solutions, we need to show that any linear program that has some solution has a basic feasible solution. This is clear geometrically: any time you have an optimum it has to either lie on a line or at a vertex, and if it lies on a line then you can slide it to a vertex without changing its value. Nevertheless, it is a useful exercise to go through the algebra.

Theorem. Whenever a linear program is feasible and bounded, it has a basic feasible solution.

Proof. Let x be an optimal solution to the LP. If x has at most m nonzero entries then it’s a basic solution and by the non-degeneracy assumption it must have exactly m nonzero entries. In this case there’s nothing to do, so suppose that x has r > m nonzero entries. It can’t be a basic feasible solution, and hence is not an extreme point of the set of feasible solutions (as proved by the last theorem). So write it as x = \delta y + (1-\delta) z for some feasible y \neq z and 0 < \delta < 1.

The only thing we know about x is it’s optimal. Let c be the cost vector, and the optimality says that \langle c,x \rangle \geq \langle c,y \rangle, and \langle c,x \rangle \geq \langle c,z \rangle. We claim that in fact these are equal, that y, z are both optimal as well. Indeed, say y were not optimal, then

\displaystyle \langle c, y \rangle < \langle c,x \rangle = \delta \langle c,y \rangle + (1-\delta) \langle c,z \rangle

Which can be rearranged to show that \langle c,y \rangle < \langle c, z \rangle. Unfortunately for x, this implies that it was not optimal all along:

\displaystyle \langle c,x \rangle < \delta \langle c, z \rangle + (1-\delta) \langle c,z \rangle = \langle c,z \rangle

An identical argument works to show z is optimal, too. Now we claim we can use y,z to get a new solution that has fewer than r nonzero entries. Once we show this we’re done: inductively repeat the argument with the smaller solution until we get down to exactly m nonzero variables. As before we know that y,z must have at least as many zeros as x. If they have more zeros we’re done. And if they have exactly as many zeros we can do the following trick. Write w = \gamma y + (1- \gamma)z for a \gamma \in \mathbb{R} we’ll choose later. Note that no matter the \gamma, w is optimal. Rewriting w = z + \gamma (y-z), we just have to pick a \gamma that ensures one of the nonzero coefficients of z is zeroed out while maintaining nonnegativity. Indeed, we can just look at the index i which minimizes z_i / (y-z)_i and use \delta = - z_i / (y-z)_i. \square.

So we have an immediate (and inefficient) combinatorial algorithm: enumerate all subsets of size m, compute the corresponding basic feasible solution x_B = A_B^{-1}b, and see which gives the biggest objective value. The problem is that, even if we knew the value of m, this would take time n^m, and it’s not uncommon for m to be in the tens or hundreds (and if we don’t know m the trivial search is exponential).

So we have to be smarter, and this is where the simplex tableau comes in.

The simplex tableau

Now say you have any basis B and any feasible solution x. For now x might not be a basic solution, and even if it is, its basis of nonzero entries might not be the same as B. We can decompose the equation Ax = b into the basis part and the non basis part:

A_Bx_B + A_{B'} x_{B'} = b

and solving the equation for x_B gives

x_B = A^{-1}_B(b - A_{B'} x_{B'})

It may look like we’re making a wicked abuse of notation here, but both A_Bx_B and A_{B'}x_{B'} are vectors of length m so the dimensions actually do work out. Now our feasible solution x has to satisfy Ax = b, and the entries of x are all nonnegative, so it must be that x_B \geq 0 and x_{B'} \geq 0, and by the equality above A^{-1}_B (b - A_{B'}x_{B'}) \geq 0 as well. Now let’s write the maximization objective \langle c, x \rangle by expanding it first in terms of the x_B, x_{B'}, and then expanding x_B.

\displaystyle \begin{aligned} \langle c, x \rangle & = \langle c_B, x_B \rangle + \langle c_{B'}, x_{B'} \rangle \\ & = \langle c_B, A^{-1}_B(b - A_{B'}x_{B'}) \rangle + \langle c_{B'}, x_{B'} \rangle \\ & = \langle c_B, A^{-1}_Bb \rangle + \langle c_{B'} - (A^{-1}_B A_{B'})^T c_B, x_{B'} \rangle \end{aligned}

If we want to maximize the objective, we can just maximize this last line. There are two cases. In the first, the vector c_{B'} - (A^{-1}_B A_{B'})^T c_B \leq 0 and A_B^{-1}b \geq 0. In the above equation, this tells us that making any component of x_{B'} bigger will decrease the overall objective. In other words, \langle c, x \rangle \leq \langle c_B, A_B^{-1}b \rangle. Picking x = A_B^{-1}b (with zeros in the non basis part) meets this bound and hence must be optimal. In other words, no matter what basis B we’ve chosen (i.e., no matter the candidate basic feasible solution), if the two conditions hold then we’re done.

Now the crux of the algorithm is the second case: if the conditions aren’t met, we can pick a positive index of c_{B'} - (A_B^{-1}A_{B'})^Tc_B and increase the corresponding value of x_{B'} to increase the objective value. As we do this, other variables in the solution will change as well (by decreasing), and we have to stop when one of them hits zero. In doing so, this changes the basis by removing one index and adding another. In reality, we’ll figure out how much to increase ahead of time, and the change will correspond to a single elementary row-operation in a matrix.

Indeed, the matrix we’ll use to represent all of this data is called a tableau in the literature. The columns of the tableau will correspond to variables, and the rows to constraints. The last row of the tableau will maintain a candidate solution y to the dual problem. Here’s a rough picture to keep the different parts clear while we go through the details.


But to make it work we do a slick trick, which is to “left-multiply everything” by A_B^{-1}. In particular, if we have an LP given by c, A, b, then for any basis it’s equivalent to the LP given by c, A_B^{-1}A, A_{B}^{-1} b (just multiply your solution to the new program by A_B to get a solution to the old one). And so the actual tableau will be of this form.


When we say it’s in this form, it’s really only true up to rearranging columns. This is because the chosen basis will always be represented by an identity matrix (as it is to start with), so to find the basis you can find the embedded identity sub-matrix. In fact, the beginning of the simplex algorithm will have the initial basis sitting in the last few columns of the tableau.

Let’s look a little bit closer at the last row. The first portion is zero because A_B^{-1}A_B is the identity. But furthermore with this A_B^{-1} trick the dual LP involves A_B^{-1} everywhere there’s a variable. In particular, joining all but the last column of the last row of the tableau, we have the vector c - A_B^T(A_B^{-1})^T c, and setting y = A_B^{-1}c_B we get a candidate solution for the dual. What makes the trick even slicker is that A_B^{-1}b is already the candidate solution x_B, since (A_B^{-1}A)_B^{-1} is the identity. So we’re implicitly keeping track of two solutions here, one for the primal LP, given by the last column of the tableau, and one for the dual, contained in the last row of the tableau.

I told you the last row was the dual solution, so why all the other crap there? This is the final slick in the trick: the last row further encodes the complementary slackness conditions. Now that we recognize the dual candidate sitting there, the complementary slackness conditions simply ask for the last row to be non-positive (this is just another way of saying what we said at the beginning of this section!). You should check this, but it gives us a stopping criterion: if the last row is non-positive then stop and output the last column.

The simplex algorithm

Now (finally!) we can describe and implement the simplex algorithm in its full glory. Recall that our informal setup has been:

  1. Find an initial basic feasible solution, and set up the corresponding tableau.
  2. Find a positive index of the last row, and increase the corresponding variable (adding it to the basis) just enough to make another variable from the basis zero (removing it from the basis).
  3. Repeat step 2 until the last row is nonpositive.
  4. Output the last column.

This is almost correct, except for some details about how increasing the corresponding variables works. What we’ll really do is represent the basis variables as pivots (ones in the tableau) and then the first 1 in each row will be the variable whose value is given by the entry in the last column of that row. So, for example, the last entry in the first row may be the optimal value for x_5, if the fifth column is the first entry in row 1 to have a 1.

As we describe the algorithm, we’ll illustrate it running on a simple example. In doing this we’ll see what all the different parts of the tableau correspond to from the previous section in each step of the algorithm.


Spoiler alert: the optimum is x_1 = 2, x_2 = 1 and the value of the max is 8.

So let’s be more programmatically formal about this. The main routine is essentially pseudocode, and the difficulty is in implementing the helper functions

def simplex(c, A, b):
   tableau = initialTableau(c, A, b)

   while canImprove(tableau):
      pivot = findPivotIndex(tableau)
      pivotAbout(tableau, pivot)

   return primalSolution(tableau), objectiveValue(tableau)

Let’s start with the initial tableau. We’ll assume the user’s inputs already include the slack variables. In particular, our example data before adding slack is

c = [3, 2]
A = [[1, 2], [1, -1]]
b = [4, 1]

And after adding slack:

c = [3, 2, 0, 0]
A = [[1,  2,  1,  0],
     [1, -1,  0,  1]]
b = [4, 1]

Now to set up the initial tableau we need an initial feasible solution in mind. The reader is recommended to work this part out with a pencil, since it’s much easier to write down than it is to explain. Since we introduced slack variables, our initial feasible solution (basis) B can just be (0,0,1,1). And so x_B is just the slack variables, c_B is the zero vector, and A_B is the 2×2 identity matrix. Now A_B^{-1}A_{B'} = A_{B'}, which is just the original two columns of A we started with, and A_B^{-1}b = b. For the last row, c_B is zero so the part under A_B^{-1}A_B is the zero vector. The part under A_B^{-1}A_{B'} is just c_{B'} = (3,2).

Rather than move columns around every time the basis B changes, we’ll keep the tableau columns in order of (x_1, \dots, x_n, \xi_1, \dots, \xi_m). In other words, for our example the initial tableau should look like this.

[[ 1,  2,  1,  0,  4],
 [ 1, -1,  0,  1,  1],
 [ 3,  2,  0,  0,  0]]

So implementing initialTableau is just a matter of putting the data in the right place.

def initialTableau(c, A, b):
   tableau = [row[:] + [x] for row, x in zip(A, b)]
   tableau.append(c[:] + [0])
   return tableau

As an aside: in the event that we don’t start with the trivial basic feasible solution of “trivially use the slack variables,” we’d have to do a lot more work in this function. Next, the primalSolution() and objectiveValue() functions are simple, because they just extract the encoded information out from the tableau (some helper functions are omitted for brevity).

def primalSolution(tableau):
   # the pivot columns denote which variables are used
   columns = transpose(tableau)
   indices = [j for j, col in enumerate(columns[:-1]) if isPivotCol(col)]
   return list(zip(indices, columns[-1]))

def objectiveValue(tableau):
   return -(tableau[-1][-1])

Similarly, the canImprove() function just checks if there’s a nonnegative entry in the last row

def canImprove(tableau):
   lastRow = tableau[-1]
   return any(x > 0 for x in lastRow[:-1])

Let’s run the first loop of our simplex algorithm. The first step is checking to see if anything can be improved (in our example it can). Then we have to find a pivot entry in the tableau. This part includes some edge-case checking, but if the edge cases aren’t a problem then the strategy is simple: find a positive entry corresponding to some entry j of B', and then pick an appropriate entry in that column to use as the pivot. Pivoting increases the value of x_j (from zero) to whatever is the largest we can make it without making some other variables become negative. As we’ve said before, we’ll stop increasing x_j when some other variable hits zero, and we can compute which will be the first to do so by looking at the current values of x_B = A_B^{-1}b (in the last column of the tableau), and seeing how pivoting will affect them. If you stare at it for long enough, it becomes clear that the first variable to hit zero will be the entry x_i of the basis for which x_i / A_{i,j} is minimal (and A_{i,j} has to be positve). This is because, in order to maintain the linear equalities, every entry of x_B will be decreased by that value during a pivot, and we can’t let any of the variables become negative.

All of this results in the following function, where we have left out the degeneracy/unboundedness checks.

[UPDATE 2018-04-21]: The pivot choices are not as simple as I thought at the time I wrote this. See the discussion on this issue, but the short story is that I was increasing the variable too much, and to fix it it’s easier to update the pivot column choice to be the smallest positive entry of the last row. The code on github is updated to reflect that, but this post will remain unchanged.

def findPivotIndex(tableau):
   # pick first nonzero index of the last row
   column = [i for i,x in enumerate(tableau[-1][:-1]) if x > 0][0]
   quotients = [(i, r[-1] / r[column]) for i,r in enumerate(tableau[:-1]) if r[column] > 0]

   # pick row index minimizing the quotient
   row = min(quotients, key=lambda x: x[1])[0]
   return row, column

For our example, the minimizer is the (1,0) entry (second row, first column). Pivoting is just doing the usual elementary row operations (we covered this in a primer a while back on row-reduction). The pivot function we use here is no different, and in particular mutates the list in place.

def pivotAbout(tableau, pivot):
   i,j = pivot

   pivotDenom = tableau[i][j]
   tableau[i] = [x / pivotDenom for x in tableau[i]]

   for k,row in enumerate(tableau):
      if k != i:
         pivotRowMultiple = [y * tableau[k][j] for y in tableau[i]]
         tableau[k] = [x - y for x,y in zip(tableau[k], pivotRowMultiple)]

And in our example pivoting around the chosen entry gives the new tableau.

[[ 0.,  3.,  1., -1.,  3.],
 [ 1., -1.,  0.,  1.,  1.],
 [ 0.,  5.,  0., -3., -3.]]

In particular, B is now (1,0,1,0), since our pivot removed the second slack variable \xi_2 from the basis. Currently our solution has x_1 = 1, \xi_1 = 3. Notice how the identity submatrix is still sitting in there, the columns are just swapped around.

There’s still a positive entry in the bottom row, so let’s continue. The next pivot is (0,1), and pivoting around that entry gives the following tableau:

[[ 0.        ,  1.        ,  0.33333333, -0.33333333,  1.        ],
 [ 1.        ,  0.        ,  0.33333333,  0.66666667,  2.        ],
 [ 0.        ,  0.        , -1.66666667, -1.33333333, -8.        ]]

And because all of the entries in the bottom row are negative, we’re done. We read off the solution as we described, so that the first variable is 2 and the second is 1, and the objective value is the opposite of the bottom right entry, 8.

To see all of the source code, including the edge-case-checking we left out of this post, see the Github repository for this post.

Obvious questions and sad answers

An obvious question is: what is the runtime of the simplex algorithm? Is it polynomial in the size of the tableau? Is it even guaranteed to stop at some point? The surprising truth is that nobody knows the answer to all of these questions! Originally (in the 1940’s) the simplex algorithm actually had an exponential runtime in the worst case, though this was not known until 1972. And indeed, to this day while some variations are known to terminate, no variation is known to have polynomial runtime in the worst case. Some of the choices we made in our implementation (for example, picking the first column with a positive entry in the bottom row) have the potential to cycle, i.e., variables leave and enter the basis without changing the objective at all. Doing something like picking a random positive column, or picking the column which will increase the objective value by the largest amount are alternatives. Unfortunately, every single pivot-picking rule is known to give rise to exponential-time simplex algorithms in the worst case (in fact, this was discovered as recently as 2011!). So it remains open whether there is a variant of the simplex method that runs in guaranteed polynomial time.

But then, in a stunning turn of events, Leonid Khachiyan proved in the 70’s that in fact linear programs can always be solved in polynomial time, via a completely different algorithm called the ellipsoid method. Following that was a method called the interior point method, which is significantly more efficient. Both of these algorithms generalize to problems that are harder than linear programming as well, so we will probably cover them in the distant future of this blog.

Despite the celebratory nature of these two results, people still use the simplex algorithm for industrial applications of linear programming. The reason is that it’s much faster in practice, and much simpler to implement and experiment with.

The next obvious question has to do with the poignant observation that whole numbers are great. That is, you often want the solution to your problem to involve integers, and not real numbers. But adding the constraint that the variables in a linear program need to be integer valued (even just 0-1 valued!) is NP-complete. This problem is called integer linear programming, or just integer programming (IP). So we can’t hope to solve IP, and rightly so: the reader can verify easily that boolean satisfiability instances can be written as linear programs where each clause corresponds to a constraint.

This brings up a very interesting theoretical issue: if we take an integer program and just remove the integrality constraints, and solve the resulting linear program, how far away are the two solutions? If they’re close, then we can hope to give a good approximation to the integer program by solving the linear program and somehow turning the resulting solution back into an integer solution. In fact this is a very popular technique called LP-rounding. We’ll also likely cover that on this blog at some point.

Oh there’s so much to do and so little time! Until next time.

Fixing Bugs in “Computing Homology”

A few awesome readers have posted comments in Computing Homology to the effect of, “Your code is not quite correct!” And they’re right! Despite the almost year since that post’s publication, I haven’t bothered to test it for more complicated simplicial complexes, or even the basic edge cases! When I posted it the mathematics just felt so solid to me that it had to be right (the irony is rich, I know).

As such I’m apologizing for my lack of rigor and explaining what went wrong, the fix, and giving some test cases. As of the publishing of this post, the Github repository for Computing Homology has been updated with the correct code, and some more examples.

The main subroutine was the simultaneousReduce function which I’ll post in its incorrectness below

def simultaneousReduce(A, B):
   if A.shape[1] != B.shape[0]:
      raise Exception("Matrices have the wrong shape.")

   numRows, numCols = A.shape # col reduce A

   i,j = 0,0
   while True:
      if i >= numRows or j >= numCols:

      if A[i][j] == 0:
         nonzeroCol = j
         while nonzeroCol < numCols and A[i,nonzeroCol] == 0:
            nonzeroCol += 1

         if nonzeroCol == numCols:
            j += 1

         colSwap(A, j, nonzeroCol)
         rowSwap(B, j, nonzeroCol)

      pivot = A[i,j]
      scaleCol(A, j, 1.0 / pivot)
      scaleRow(B, j, 1.0 / pivot)

      for otherCol in range(0, numCols):
         if otherCol == j:
         if A[i, otherCol] != 0:
            scaleAmt = -A[i, otherCol]
            colCombine(A, otherCol, j, scaleAmt)
            rowCombine(B, j, otherCol, -scaleAmt)

      i += 1; j+= 1

   return A,B

It’s a beast of a function, and the persnickety detail was just as beastly: this snippet should have an i += 1 instead of a j.

if nonzeroCol == numCols:
   j += 1

This is simply what happens when we’re looking for a nonzero entry in a row to use as a pivot for the corresponding column, but we can’t find one and have to move to the next row. A stupid error on my part that would be easily caught by proper test cases.

The next mistake is a mathematical misunderstanding. In short, the simultaneous column/row reduction process is not enough to get the \partial_{k+1} matrix into the right form! Let’s see this with a nice example, a triangulation of the Mobius band. There are a number of triangulations we could use, many of which are seen in these slides. The one we’ll use is the following.


It’s first and second boundary maps are as follows (in code, because latex takes too much time to type out)

mobiusD1 = numpy.array([
   [-1,-1,-1,-1, 0, 0, 0, 0, 0, 0],
   [ 1, 0, 0, 0,-1,-1,-1, 0, 0, 0],
   [ 0, 1, 0, 0, 1, 0, 0,-1,-1, 0],
   [ 0, 0, 0, 1, 0, 0, 1, 0, 1, 1],

mobiusD2 = numpy.array([
   [ 1, 0, 0, 0, 1],
   [ 0, 0, 0, 1, 0],
   [-1, 0, 0, 0, 0],
   [ 0, 0, 0,-1,-1],
   [ 0, 1, 0, 0, 0],
   [ 1,-1, 0, 0, 0],
   [ 0, 0, 0, 0, 1],
   [ 0, 1, 1, 0, 0],
   [ 0, 0,-1, 1, 0],
   [ 0, 0, 1, 0, 0],

And if we were to run the above code on it we’d get a first Betti number of zero (which is incorrect, it’s first homology group has rank 1). Here are the reduced matrices.

>>> A1, B1 = simultaneousReduce(mobiusD1, mobiusD2)
>>> A1
array([[1, 0, 0, 0, 0, 0, 0, 0, 0, 0],
       [0, 1, 0, 0, 0, 0, 0, 0, 0, 0],
       [0, 0, 1, 0, 0, 0, 0, 0, 0, 0],
       [0, 0, 0, 1, 0, 0, 0, 0, 0, 0]])
>>> B1
array([[ 0,  0,  0,  0,  0],
       [ 0,  0,  0,  0,  0],
       [ 0,  0,  0,  0,  0],
       [ 0,  0,  0,  0,  0],
       [ 0,  1,  0,  0,  0],
       [ 1, -1,  0,  0,  0],
       [ 0,  0,  0,  0,  1],
       [ 0,  1,  1,  0,  0],
       [ 0,  0, -1,  1,  0],
       [ 0,  0,  1,  0,  0]])

The first reduced matrix looks fine; there’s nothing we can do to improve it. But the second one is not quite fully reduced! Notice that rows 5, 8 and 10 are not linearly independent. So we need to further row-reduce the nonzero part of this matrix before we can read off the true rank in the way we described last time. This isn’t so hard (we just need to reuse the old row-reduce function we’ve been using), but why is this allowed? It’s just because the corresponding column operations for those row operations are operating on columns of all zeros! So we need not worry about screwing up the work we did in column reducing the first matrix, as long as we only work with the nonzero rows of the second.

Of course, nothing is stopping us from ignoring the “corresponding” column operations, since we know we’re already done there. So we just have to finish row reducing this matrix.

This changes our bettiNumber function by adding a single call to a row-reduce function which we name so as to be clear what’s happening. The resulting function is

def bettiNumber(d_k, d_kplus1):
   A, B = numpy.copy(d_k), numpy.copy(d_kplus1)
   simultaneousReduce(A, B)

   dimKChains = A.shape[1]
   kernelDim = dimKChains - numPivotCols(A)
   imageDim = numPivotRows(B)

   return kernelDim - imageDim

And running this on our Mobius band example gives:

>>> bettiNumber(mobiusD1, mobiusD2))

As desired. Just to make sure things are going swimmingly under the hood, we can check to see how finishRowReducing does after calling simultaneousReduce

>>> simultaneousReduce(mobiusD1, mobiusD2)
(array([[1, 0, 0, 0, 0, 0, 0, 0, 0, 0],
       [0, 1, 0, 0, 0, 0, 0, 0, 0, 0],
       [0, 0, 1, 0, 0, 0, 0, 0, 0, 0],
       [0, 0, 0, 1, 0, 0, 0, 0, 0, 0]]), array([[ 0,  0,  0,  0,  0],
       [ 0,  0,  0,  0,  0],
       [ 0,  0,  0,  0,  0],
       [ 0,  0,  0,  0,  0],
       [ 0,  1,  0,  0,  0],
       [ 1, -1,  0,  0,  0],
       [ 0,  0,  0,  0,  1],
       [ 0,  1,  1,  0,  0],
       [ 0,  0, -1,  1,  0],
       [ 0,  0,  1,  0,  0]]))
>>> finishRowReducing(mobiusD2)
array([[1, 0, 0, 0, 0],
       [0, 1, 0, 0, 0],
       [0, 0, 1, 0, 0],
       [0, 0, 0, 1, 0],
       [0, 0, 0, 0, 1],
       [0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0]])

Indeed, finishRowReducing finishes row reducing the second boundary matrix. Note that it doesn’t preserve how the rows of zeros lined up with the pivot columns of the reduced version of \partial_1 as it did in the previous post, but since in the end we’re only counting pivots it doesn’t matter how we switch rows. The “zeros lining up” part is just for a conceptual understanding of how the image lines up with the kernel for a valid simplicial complex.

In fixing this issue we’ve also fixed an issue another commenter mentioned, that you couldn’t blindly plug in the zero matrix for \partial_0 and get zeroth homology (which is the same thing as connected components). After our fix you can.

Of course there still might be bugs, but I have so many drafts lined up on this blog (and research papers to write, experiments to run, theorems to prove), that I’m going to put off writing a full test suite. I’ll just have to update this post with new bug fixes as they come. There’s just so much math and so little time 🙂 But extra kudos to my amazing readers who were diligent enough to run examples and spot my error. I’m truly blessed to have you on my side.

Also note that this isn’t the most efficient way to represent the simplicial complex data, or the most efficient row reduction algorithm. If you’re going to run the code on big inputs, I suggest you take advantage of sparse matrix algorithms for doing this sort of stuff. You can represent the simplices as entries in a dictionary and do all sorts of clever optimizations to make the algorithm effectively linear time in the number of simplices.

Until next time!

Homology Theory — A Primer

This series on topology has been long and hard, but we’re are quickly approaching the topics where we can actually write programs. For this and the next post on homology, the most important background we will need is a solid foundation in linear algebra, specifically in row-reducing matrices (and the interpretation of row-reduction as a change of basis of a linear operator).

Last time we engaged in a whirlwind tour of the fundamental group and homotopy theory. And we mean “whirlwind” as it sounds; it was all over the place in terms of organization. The most important fact that one should take away from that discussion is the idea that we can compute, algebraically, some qualitative features about a topological space related to “n-dimensional holes.” For one-dimensional things, a hole would look like a circle, and for two dimensional things, it would look like a hollow sphere, etc. More importantly, we saw that this algebraic data, which we called the fundamental group, is a topological invariant. That is, if two topological spaces have different fundamental groups, then they are “fundamentally” different under the topological lens (they are not homeomorphic, and not even homotopy equivalent).

Unfortunately the main difficulty of homotopy theory (and part of what makes it so interesting) is that these “holes” interact with each other in elusive and convoluted ways, and the algebra reflects it almost too well. Part of the problem with the fundamental group is that it deftly eludes our domain of interest: computing them is complicated!

What we really need is a coarser invariant. If we can find a “stupider” invariant, it might just be simple enough to compute efficiently. Perhaps unsurprisingly, these will take the form of finitely-generated abelian groups (the most well-understood class of groups), with one for each dimension. Now we’re starting to see exactly why algebraic topology is so difficult; it has an immense list of prerequisite topics! If we’re willing to skip over some of the more nitty gritty details (and we must lest we take a huge diversion to discuss Tor and the exact sequences in the universal coefficient theorem), then we can also do the same calculations over a field. In other words, the algebraic objects we’ll define called “homology groups” are really vector spaces, and so row-reduction will be our basic computational tool to analyze them.

Once we have the basic theory down, we’ll see how we can write a program which accepts as input any topological space (represented in a particular form) and produces as output a list of the homology groups in every dimension. The dimensions of these vector spaces (their ranks, as finitely-generated abelian groups) are interpreted as the number of holes in the space for each dimension.

Recall Simplicial Complexes

In our post on constructing topological spaces, we defined the standard k-simplex and the simplicial complex. We recall the latter definition here, and expand upon it.

Definition: A simplicial complex is a topological space realized as a union of any collection of simplices (of possibly varying dimension) \Sigma which has the following two properties:

  • Any face of a simplex \Sigma is also in \Sigma.
  • The intersection of any two simplices of \Sigma is also a simplex of \Sigma.

We can realize a simplicial complex by gluing together pieces of increasing dimension. First start by taking a collection of vertices (0-simplices) X_0. Then take a collection of intervals (1-simplices) X_1 and glue their endpoints onto the vertices in any way. Note that because we require every face of an interval to again be a simplex in our complex, we must glue each endpoint of an interval onto a vertex in X_0. Continue this process with X_2, a set of 2-simplices, we must glue each edge precisely along an edge of X_1. We can continue this process until we reach a terminating set X_n. It is easy to see that the union of the X_i form a simplicial complex. Define the dimension of the cell complex to be n.

There are some picky restrictions on how we glue things that we should mention. For instance, we could not contract all edges of a 2-simplex \sigma and glue it all to a single vertex in X_0. The reason for this is that \sigma would no longer be a 2-simplex! Indeed, we’ve destroyed its original vertex set. The gluing process hence needs to preserve the original simplex’s boundary. Moreover, one property that follows from the two conditions above is that any simplex in the complex is uniquely determined by its vertices (for otherwise, the intersection of two such non-uniquely specified simplices would not be a single simplex).

We also have to remember that we’re imposing a specific ordering on the vertices of a simplex. In particular, if we label the vertices of an n-simplex 0, \dots, n, then this imposes an orientation on the edges where an edge of the form \left \{ i,j \right \} has the orientation (i,j) if i < j, and (j,i) otherwise. The faces, then, are “oriented” in increasing order of their three vertices. Higher dimensional simplices are oriented in a similar way, though we rarely try to picture this (the theory of orientations is a question best posted for smooth manifolds; we won’t be going there any time soon). Here are, for example, two different ways to pick orientations of a 2-simplex:

Two possible orientations of a 2-simplex

Two possible orientations of a 2-simplex.

It is true, but a somewhat lengthy exercise, that the topology of a simplicial complex does not change under a consistent shuffling of the orientations across all its simplices. Nor does it change depending on how we realize a space as a simplicial complex. These kinds of results are crucial to the welfare of the theory, but have been proved once and we won’t bother reproving them here.

As a larger example, here is a simplicial complex representing the torus. It’s quite a bit more complicated than our usual quotient of a square, but it’s based on the same idea. The left and right edges are glued together, as are the top and bottom, with appropriate orientations. The only difficulty is that we need each simplex to be uniquely determined by its vertices. While this construction does not use the smallest possible number of simplices to satisfy that condition, it is the simplest to think about.

A possible realization of the torus as a simplicial complex. As an exercise, the reader is invited to fill in the orientations on the simplices to be consistent across the entire complex.

A possible realization of the torus as a simplicial complex. As an exercise, the reader is invited to label the edges and fill in the orientations on the simplices to be consistent across the entire complex. Remember that the result should coincide with our classical construction via the quotient of the disk, so some of the edges on the sides will coincide with those on the opposite sides, and the orientations must line up.

Taking a known topological space (like the torus) and realizing it as a simplicial complex is known as triangulating the space. A space which can be realized as a simplicial complex is called triangulable.

The nicest thing about the simplex is that it has an easy-to-describe boundary. Geometrically, it’s obvious: the boundary of the line segment is the two endpoints; the boundary of the triangle is the union of all three of its edges; the tetrahedron has four triangular faces as its boundary; etc. But because we need an algebraic way to describe holes, we want an algebraic way to describe the boundary. In particular, we have two important criterion that any algebraic definition must satisfy to be reasonable:

  1. A boundary itself has no boundary.
  2. The property of being boundariless (at least in low dimensions) coincides with our intuitive idea of what it means to be a loop.

Of course, just as with homotopy these holes interact in ways we’re about to see, so we need to be totally concrete before we can proceed.

The Chain Group and the Boundary Operator

In order to define an algebraic boundary, we have to realize simplices themselves as algebraic objects.  This is not so difficult to do: just take all “formal sums” of simplices in the complex. More rigorously, let X_k be the set of k-simplices in the simplicial complex X. Define the chain group C_k(X) to be the \mathbb{Q}-vector space with X_k for a basis. The elements of the k-th chain group are called k-chainon X. That’s right, if \sigma, \sigma' are two k-simplices, then we just blindly define a bunch of new “chains” as all possible “sums” and scalar multiples of the simplices. For example, sums involving two elements would look like a\sigma + b\sigma' for some a,b \in \mathbb{Q}. Indeed, we include any finite sum of such simplices, as is standard in taking the span of a set of basis vectors in linear algebra.

Just for a quick example, take this very simple simplicial complex:


We’ve labeled all of the simplices above, and we can describe the chain groups quite easily. The zero-th chain group C_0(X) is the \mathbb{Q}-linear span of the set of vertices \left \{ v_1, v_2, v_3, v_4 \right \}. Geometrically, we might think of “the union” of two points as being, e.g., the sum v_1 + v_3. And if we want to have two copies of v_1 and five copies of v_3, that might be thought of as 2v_1 + 5v_3. Of course, there are geometrically meaningless sums like \frac{1}{2}v_4 - v_2 - \frac{11}{6}v_1, but it will turn out that the algebra we use to talk about holes will not falter because of it. It’s nice to have this geometric idea of what an algebraic expression can “mean,” but in light of this nonsense it’s not a good idea to get too wedded to the interpretations.

Likewise, C_1(X) is the linear span of the set \left \{ e_1, e_2, e_3, e_4, e_5 \right \} with coefficients in \mathbb{Q}. So we can talk about a “path” as a sum of simplices like e_1 + e_4 - e_5 + e_3. Here we use a negative coefficient to signify that we’re travelling “against” the orientation of an edge. Note that since the order of the terms is irrelevant, the same “path” is given by, e.g. -e_5 + e_4 + e_1 + e_3, which geometrically is ridiculous if we insist on reading the terms from left to right.

The same idea extends to higher dimensional groups, but as usual the visualization grows difficult. For example, in C_2(X) above, the chain group is the vector space spanned by \left \{ \sigma_1, \sigma_2 \right \}. But does it make sense to have a path of triangles? Perhaps, but the geometric analogies certainly become more tenuous as dimension grows. The benefit, however, is if we come up with good algebraic definitions for the low-dimensional cases, the algebra is easy to generalize to higher dimensions.

So now we will define the boundary operator on chain groups, a linear map \partial : C_k(X) \to C_{k-1}(X) by starting in lower dimensions and generalizing. A single vertex should always be boundariless, so \partial v = 0 for each vertex. Extending linearly to the entire chain group, we have \partial is identically the zero map on zero-chains. For 1-simplices we have a more substantial definition: if a simplex has its orientation as (v_1, v_2), then the boundary \partial (v_1, v_2) should be v_2 - v_1. That is, it’s the front end of the edge minus the back end. This defines the boundary operator on the basis elements, and we can again extend linearly to the entire group of 1-chains.

Why is this definition more sensible than, say, v_1 + v_2? Using our example above, let’s see how it operates on a “path.” If we have a sum like e_1 + e_4 - e_5 - e_3, then the boundary is computed as

\displaystyle \partial (e_1 + e_4 - e_5 - e_3) = \partial e_1 + \partial e_4 - \partial e_5 - \partial e_3
\displaystyle = (v_2 - v_1) + (v_4 - v_2) - (v_4 - v_3) - (v_3 - v_2) = v_2 - v_1

That is, the result was the endpoint of our path v_2 minus the starting point of our path v_1. It is not hard to prove that this will work in general, since each successive edge in a path will cancel out the ending vertex of the edge before it and the starting vertex of the edge after it: the result is just one big alternating sum.

Even more importantly is that if the “path” is a loop (the starting and ending points are the same in our naive way to write the paths), then the boundary is zero. Indeed, any time the boundary is zero then one can rewrite the sum as a sum of “loops,” (though one might have to trivially introduce cancelling factors). And so our condition for a chain to be a “loop,” which is just one step away from being a “hole,” is if it is in the kernel of the boundary operator. We have a special name for such chains: they are called cycles.

For 2-simplices, the definition is not so much harder: if we have a simplex like (v_0, v_1, v_2), then the boundary should be (v_1,v_2) - (v_0,v_2) + (v_0,v_1). If one rewrites this in a different order, then it will become apparent that this is just a path traversing the boundary of the simplex with the appropriate orientations. We wrote it in this “backwards” way to lead into the general definition: the simplices are ordered by which vertex does not occur in the face in question (v_0 omitted from the first, v_1 from the second, and v_2 from the third).

We are now ready to extend this definition to arbitrary simplices, but a nice-looking definition requires a bit more notation. Say we have a k-simplex which looks like (v_0, v_1, \dots, v_k). Abstractly, we can write it just using the numbers, as [0,1,\dots, k]. And moreover, we can denote the removal of a vertex from this list by putting a hat over the removed index. So [0,1,\dots, \hat{i}, \dots, k] represents the simplex which has all of the vertices from 0 to k excluding the vertex v_i. To represent a single-vertex simplex, we will often drop the square brackets, e.g. 3 for [3]. This can make for some awkward looking math, but is actually standard notation once the correct context has been established.

Now the boundary operator is defined on the standard n-simplex with orientation [0,1,\dots, n] via the alternating sum

\displaystyle \partial([0,1,\dots, n]) = \sum_{k=0}^n (-1)^k [0, \dots, \hat{k}, \dots, n]

It is trivial (but perhaps notationally hard to parse) to see that this coincides with our low-dimensional examples above. But now that we’ve defined it for the basis elements of a chain group, we automatically get a linear operator on the entire chain group by extending \partial linearly on chains.

Definition: The k-cycles on X are those chains in the kernel of \partial. We will call k-cycles boundariless. The k-boundaries are the image of \partial.

We should note that we are making a serious abuse of notation here, since technically \partial is defined on only a single chain group. What we should do is define \partial_k : C_k(X) \to C_{k-1}(X) for a fixed dimension, and always put the subscript. In practice this is only done when it is crucial to be clear which dimension is being talked about, and otherwise the dimension is always inferred from the context. If we want to compose the boundary operator in one dimension with the boundary operator in another dimension (say, \partial_{k-1} \partial_k), it is usually written \partial^2. This author personally supports the abolition of the subscripts for the boundary map, because subscripts are a nuisance in algebraic topology.

All of that notation discussion is so we can make the following observation: \partial^2 = 0. That is, every chain which is a boundary of a higher-dimensional chain is boundariless! This should make sense in low-dimension: if we take the boundary of a 2-simplex, we get a cycle of three 1-simplices, and the boundary of this chain is zero. Indeed, we can formally prove it from the definition for general simplices (and extend linearly to achieve the result for all simplices) by writing out \partial^2([0,1,\dots, n]). With a keen eye, the reader will notice that the terms cancel out and we get zero. The reason is entirely in which coefficients are negative; the second time we apply the boundary operator the power on (-1) shifts by one index. We will leave the full details as an exercise to the reader.

So this fits our two criteria: low-dimensional examples make sense, and boundariless things (cycles) represent loops.

Recasting in Algebraic Terms, and the Homology Group

For the moment let’s give boundary operators subscripts \partial_k : C_k(X) \to C_{k-1}(X). If we recast things in algebraic terms, we can call the k-cycles Z_k(X) = \textup{ker}(\partial_k), and this will be a subspace (and a subgroup) of C_k(X) since kernels are always linear subspaces. Moreover, the set B_k(X) of k-boundaries, that is, the image of \partial_{k+1}, is a subspace (subgroup) of Z_k(X). As we just saw, every boundary is itself boundariless, so B_k(X) is a subset of Z_k(X), and since the image of a linear map is always a linear subspace of the range, we get that it is a subspace too.

All of this data is usually expressed in one big diagram: each of the chain groups are organized in order of decreasing dimension, and the boundary maps connect them.


Since our example (the “simple space” of two triangles from the previous section) only has simplices in dimensions zero, one, and two, we additionally extend the sequence of groups to an infinite sequence by adding trivial groups and zero maps, as indicated. The condition that \textup{im} \partial_{k+1} \subset \textup{ker} \partial_k, which is equivalent to \partial^2 = 0, is what makes this sequence a chain complex. As a side note, every sequence of abelian groups and group homomorphisms which satisfies the boundary requirement is called an algebraic chain complex. This foreshadows that there are many different types of homology theory, and they are unified by these kinds of algebraic conditions.

Now, geometrically we want to say, “The holes are all those cycles (loops) which don’t arise as the boundaries of higher-dimensional things.” In algebraic terms, this would correspond to a quotient space (really, a quotient group, which we covered in our first primer on groups) of the k-cycles by the k-boundaries. That is, a cycle would be considered a “trivial hole” if it is a boundary, and two “different” cycles would be considered the same hole if their difference is a k-boundary. This is the spirit of homology, and formally, we define the homology group (vector space) as follows.

Definition: The k-th homology group of a simplicial complex X, denoted H_k(X), is the quotient vector space Z_k(X) / B_k(X) = \textup{ker}(\partial_k) / \textup{im}(\partial_{k+1}). Two elements of a homology group which are equivalent (their difference is a boundary) are called homologous.

The number of k-dimensional holes in X is thus realized as the dimension of H_k(X) as a vector space.

The quotient mechanism really is doing all of the work for us here. Any time we have two holes and we’re wondering whether they represent truly different holes in the space (perhaps we have a closed loop of edges, and another which is slightly longer but does not quite use the same edges), we can determine this by taking their difference and seeing if it bounds a higher-dimensional chain. If it does, then the two chains are the same, and if it doesn’t then the two chains carry intrinsically different topological information.

For particular dimensions, there are some neat facts (which obviously require further proof) that make this definition more concrete.

  • The dimension of H_0(X) is the number of connected components of X. Therefore, computing homology generalizes the graph-theoretic methods of computing connected components.
  • H_1(X) is the abelianization of the fundamental group of X. Roughly speaking, H_1(X) is the closest approximation of \pi_1(X) by a \mathbb{Q} vector space.

Now that we’ve defined the homology group, let’s end this post by computing all the homology groups for this example space:


This is a sphere (which can be triangulated as the boundary of a tetrahedron) with an extra “arm.” Note how the edge needs an extra vertex to maintain uniqueness. This space is a nice example because it has one-dimensional homology in dimension zero (one connected component), dimension one (the arm is like a copy of the circle), and dimension two (the hollow sphere part). Let’s verify this algebraically.

Let’s start by labelling the vertices of the tetrahedron 0, 1, 2, 3, so that the extra arm attaches at 0 and 2, and call the extra vertex on the arm 4. This completely determines the orientations for the entire simplex, as seen below.

Indeed, the chain groups are easy to write down:

\displaystyle C_0(X) = \textup{span} \left \{ 0,1,2,3,4 \right \}

\displaystyle C_1(X) = \textup{span} \left \{ [0,1], [0,2], [0,3], [0,4], [1,2], [1,3],[2,3],[2,4] \right \}

\displaystyle C_2(X) = \textup{span} \left \{ [0,1,2], [0,1,3], [0,2,3], [1,2,3] \right \}

We can easily write down the images of each \partial_k, they’re just the span of the images of each basis element under \partial_k.

\displaystyle \textup{im} \partial_1 = \textup{span} \left \{ 1 - 0, 2 - 0, 3 - 0, 4 - 0, 2 - 1, 3 - 1, 3 - 2, 4 - 2 \right \}

The zero-th homology H_0(X) is the kernel of \partial_0 modulo the image of \partial_1. The angle brackets are a shorthand for “span.”

\displaystyle \frac{\left \langle 0,1,2,3,4 \right \rangle}{\left \langle 1-0,2-0,3-0,4-0,2-1,3-1,3-2,4-2 \right \rangle}

Since \partial_0 is actually the zero map, Z_0(X) = C_0(X) and all five vertices generate the kernel. The quotient construction imposes that two vertices (two elements of the homology group) are considered equivalent if their difference is a boundary. It is easy to see that (indeed, just by the first four generators of the image) all vertices are equivalent to 0, so there is a unique generator of homology, and the vector space is isomorphic to \mathbb{Q}. There is exactly one connected component. Geometrically we can realize this, because two vertices are homologous if and only if there is a “path” of edges from one vertex to the other. This chain will indeed have as its image the difference of the two vertices.

We can compute the first homology H_1(X) in an analogous way, compute the kernel and image separately, and then compute the quotient.

\textup{ker} \partial_1 = \textup{span} \left \{ [0,1] + [0,3] - [1,3], [0,2] + [2,3] - [0,3], [1,2] + [2,3] - [1,3], [0,1] + [1,2] - [0,2], [0,2] + [2,4] - [0,4] \right \}

It takes a bit of combinatorial analysis to show that this is precisely the kernel of \partial_1, and we will have a better method for it in the next post, but indeed this is it. As the image of \partial_2 is precisely the first four basis elements, the quotient is just the one-dimensional vector space spanned by [0,2] + [2,4] - [0,4]. Hence H_1(X) = \mathbb{Q}, and there is one one-dimensional hole.

Since there are no 3-simplices, the homology group H_2(X) is simply the kernel of \partial_2, which is not hard to see is just generated by the chain representing the “sphere” part of the space: [1,2,3] - [0,2,3] + [0,1,3] - [0,1,2]. The second homology group is thus again \mathbb{Q} and there is one two-dimensional hole in X.

So there we have it!

Looking Forward

Next time, we will give a more explicit algorithm for computing homology for finite simplicial complexes, and it will essentially be a variation of row-reduction which simultaneously rewrites the matrix representations of the boundary operators \partial_{k+1}, \partial_k with respect to a canonical basis. This will allow us to simply count entries on the digaonals of the two matrices, and the difference will be the dimension of the quotient space, and hence the number of holes.

Until then!