Relative Homology

In the mathematical field of topology, homology provides a powerful method for understanding shape by counting an object's holes. But what if we are not interested in all the holes, but only those that are 'new' relative to a known part of the object? For instance, a geologist might want to study the new voids inside a block of rock, ignoring the known fissures in its outer crust. Standard homology cannot make this distinction. This gap is precisely where relative homology comes in, offering a refined lens to focus on the features of a space that are not already accounted for by a chosen subspace.

This article provides a comprehensive introduction to this fundamental concept. We will first delve into the ​​Principles and Mechanisms​​ of relative homology, exploring how it is formally defined by algebraically 'collapsing' a subspace and how the powerful long exact sequence allows us to compute it. Following this, the chapter on ​​Applications and Interdisciplinary Connections​​ will demonstrate how this seemingly abstract tool becomes a practical instrument for simplifying complex spaces, systematically building them from simple parts, and even creating invariants to study knots, revealing its deep connections across mathematics and physics.

In our journey to understand shape, we've developed a powerful tool called homology, a method for counting holes. But sometimes, counting all the holes in an object isn't quite what we want. Imagine you're a geologist studying a large block of porous rock. You might already know about the bubbles and fissures present in a certain region, say, the outer crust. Your real interest is in the new voids and channels in the interior, relative to that known crust. You want to characterize the features of the whole rock that are not already accounted for by the features of its crust. This is precisely the question that ​​relative homology​​ was invented to answer.

Let's say we have a topological space $X$ (our block of rock) and a subspace $A$ within it (the crust). The relative homology groups, denoted $H_n(X, A)$, are designed to capture the topological features of $X$ that are not already present in $A$. They measure the $n$-dimensional holes in $X$ that are formed by parts of $X$ which are not confined to $A$.

How can we make this idea precise? You might first guess that we could just "subtract" the homology of $A$ from the homology of $X$. But the holes in $A$ are also holes in $X$, and their relationship can be complicated. A loop in $A$ might be fillable in the larger space $X$, for instance. The subtraction idea is too simple; it doesn't respect the geometry. We need a much cleverer, and ultimately more beautiful, approach.

The genius of algebraic topology is to translate geometric problems into algebra. Here, the translation is wonderfully intuitive. To study $X$ relative to $A$, we perform an algebraic manipulation that is tantamount to "collapsing $A$ to a point." We decide that any part of our space that lies entirely within $A$ is uninteresting.

Formally, homology is built from objects called ​​chains​​, which are formal sums of simple geometric pieces like points, paths, and triangles. The relative homology groups $H_n(X, A)$ are computed from ​​relative chains​​, which are defined by the quotient $C_n(X, A) = C_n(X) / C_n(A)$. In this algebraic quotient, we are essentially declaring any chain that lives completely inside $A$ to be equivalent to zero. We've effectively made $A$ invisible to our hole-counting apparatus. We only "see" the chains, or parts of chains, that stick out of $A$.

Let's look at a couple of simple scenarios to see how this plays out.

• What is the homology of a space $X$ relative to itself? That is, what is $H_n(X, X)$? Here, our subspace $A$ is the entire space $X$. We've declared everything to be uninteresting! The quotient of chain groups $C_n(X)/C_n(X)$ is the trivial group, so all the resulting homology groups must be trivial. There can be no "new" features in $X$ that aren't already in $X$. This is a crucial sanity check.
• Now for something slightly more interesting. Let our space $X$ be two disconnected points, $\{p, q\}$, and let our subspace be just one of them, $A = \{p\}$. When we compute the relative homology $H_n(X, A)$, we are algebraically ignoring the point $p$. What's left? Just the point $q$. And indeed, the calculation shows that the 0-th relative homology group $H_0(X, A)$ is isomorphic to $\mathbb{Z}$, the group that signifies a single connected component (in this case, the leftover point $q$). All higher relative homology groups are zero. Relative homology has successfully isolated the part of $X$ that is not in $A$.

While the definition using chain groups is the foundation, it's often cumbersome for calculations. Fortunately, relative homology doesn't exist in a vacuum. It is intrinsically linked to the absolute homology of $X$ and $A$ through a magnificent structure known as the ​​long exact sequence of a pair​​. For any pair $(X, A)$, there exists a sequence of groups and homomorphisms that marches on infinitely in both directions:

$\dots \to H_n(A) \xrightarrow{i_*} H_n(X) \xrightarrow{j_*} H_n(X, A) \xrightarrow{\partial} H_{n-1}(A) \xrightarrow{i_*} H_{n-1}(X) \to \dots$

This sequence has a remarkable property: it is ​​exact​​. This means that at every stage, the image of the incoming map is precisely the kernel of the outgoing map. Think of it as a series of perfectly engineered pipelines and junctions. The flow of information is perfectly conserved; what flows out of one map as its image flows into the next and is completely annihilated. This exactness property turns the sequence into an incredibly powerful computational engine. If you know some of the groups in the sequence, you can often deduce the others.

This engine allows us to derive profound consequences about the nature of space.

When Nothing is New

What happens if all the relative homology groups $H_n(X, A)$ are trivial? Looking at the long exact sequence, if the group $H_n(X, A)$ is zero, the "pipelines" on either side must connect directly. Exactness forces the map $i_*: H_n(A) \to H_n(X)$ to be an isomorphism for every $n$. This means that, from the perspective of homology, $A$ and $X$ have exactly the same number and type of holes in every dimension.

This occurs in a very important geometric situation: when $A$ is a ​​deformation retract​​ of $X$. This means you can continuously shrink the entire space $X$ onto $A$ without ever tearing it and without moving the points that are already in $A$. For example, a solid disk deformation retracts onto its center point. A cylinder deformation retracts onto its central axis. If $X$ can be "squashed" down to $A$ in this way, it makes intuitive sense that they are topologically "the same," and therefore no new holes are created in $X$. The long exact sequence confirms this intuition: if $A$ is a deformation retract of $X$, then all relative homology groups $H_n(X, A)$ are zero. A great example is the pair $(\mathbb{R}^n, A)$ where $A$ is any non-empty convex set (like a point, a line, or a closed ball). The entire space $\mathbb{R}^n$ can be squashed onto $A$, so their relative homology is trivial in all dimensions.

A Universal Reference Point

A particularly illuminating case is to take the subspace $A$ to be just a single point, $\{x_0\}$. What does $H_n(X, \{x_0\})$ measure? It measures the holes in $X$ while ignoring the component containing $x_0$. This sounds very similar to another concept, ​​reduced homology​​, denoted $\tilde{H}_n(X)$, which is a slight modification of standard homology designed to make the homology of a point trivial in all dimensions. The long exact sequence provides the stunning answer: for any non-empty space $X$ and any point $x_0 \in X$, there is a natural isomorphism $H_n(X, \{x_0\}) \cong \tilde{H}_n(X)$ for all $n \ge 0$. This is a beautiful piece of unity. It tells us that relative homology is not some ad-hoc construction; it's a deep generalization that contains other fundamental ideas within it.

Revealing Hidden Twists

The most exciting applications arise when the maps in the long exact sequence are not simple. Consider the ​​Möbius strip​​, $M$, a classic one-sided surface. Its boundary, $A$, is a single circle. Both the strip itself and its boundary have the first homology group of a circle, $H_1 \cong \mathbb{Z}$. But how does the boundary circle sit inside the strip? If you trace the center line of the strip, that's one loop. If you trace the boundary circle, you'll find you have to go around the strip twice to get back to where you started.

The long exact sequence captures this perfectly. The map $i_*: H_1(A) \to H_1(M)$ induced by the inclusion is not an isomorphism, but multiplication by 2 on the integers. When we feed this into our engine, the sequence churns and spits out a remarkable result: the first relative homology group is $H_1(M, A) \cong \mathbb{Z}/2\mathbb{Z}$. This group, $\mathbb{Z}_2$, is a group with only two elements, a clear algebraic signal of the "twist" in the Möbius strip. It's a feature that doesn't belong to the boundary alone, nor is it immediately obvious from the homology of the strip alone. It is a ​​relative​​ feature, a property of the relationship between the space and its subspace, uncovered by the machinery of relative homology.

The principles we've discussed are the foundation for even more powerful tools that allow us to compute the homology of almost any space we can imagine.

• ​​Excision Theorem​​: This is a powerful "cut-and-paste" theorem. It states that if we have a pair $(X, A)$, we can cut out a suitable subset $U$ from the interior of $A$ and the relative homology won't change: $H_n(X, A) \cong H_n(X \setminus U, A \setminus U)$. This allows us to simplify spaces by excising, or removing, irrelevant parts to focus on the crucial boundary region. The proof of this theorem is technical. A core algebraic tool used elsewhere in the theory is the ​​Five-Lemma​​, which guarantees that if a map of pairs induces isomorphisms on the homology of the individual spaces and subspaces, it also does so on their relative homology.

• ​​Additivity and Triples​​: The theory is also beautifully self-consistent. For instance, if you have a space made of disjoint pieces, the relative homology is just the direct sum of the relative homologies of the pieces. Furthermore, if you have a "Russian doll" nesting of spaces $B \subset A \subset X$, there is a ​​long exact sequence of a triple​​ that relates the relative homology groups of the three pairs $(X, A)$, $(A, B)$, and $(X, B)$. This shows the deep, hierarchical structure that homology imposes on our understanding of space.

Relative homology, then, is far more than a technical modification of an existing theory. It is a lens that allows us to adjust our focus, to ignore what we already know and to bring into sharp relief the new and often subtle structures that emerge when one space is embedded inside another. It is a fundamental tool for exploring the intricate relationships that define the shape of our world.

In our last discussion, we uncovered a new kind of mathematical instrument: relative homology. We likened it to a set of colored filters, allowing us to see not just the whole picture, but to subtract a piece and examine what remains. A fair question to ask at this point is, "What on Earth is this good for?" Why would we ever want to measure the homology of 'a space minus a subspace'? It sounds like a rather abstract, if not perverse, thing to do. Yet, as is so often the case in science, by looking at something in a peculiar new way, we gain a power we never expected. Relative homology is not about destruction; it's about isolation. It's a microscope for seeing how the different parts of a complex object are glued together, and for understanding what is truly "new" when one space is built upon another.

One of the most elegant tricks that relative homology allows is a profound form of simplification. Often, a pair of spaces $(X, A)$ is difficult to visualize, but if we imagine collapsing the subspace $A$ down to a single point, the resulting quotient space $X/A$ can be something wonderfully familiar. Relative homology provides a magical bridge: for a vast class of well-behaved pairs, the relative homology of the pair is precisely the (reduced) homology of the simplified quotient space.

Let's start with the most fundamental building blocks. Consider an $n$-dimensional solid disk, $D^n$, and its boundary, the $(n-1)$-dimensional sphere $S^{n-1}$. What is the homology of the disk relative to its boundary? What are we measuring? We are asking what topological features are "inside" the disk that are not on its boundary. Intuitively, the disk is just "filled in". If we take the solid disk and shrink its entire boundary sphere to a single point, what do we get? The whole disk, with its boundary now pinched together, becomes an $n$-dimensional sphere, $S^n$. Relative homology makes this intuition precise: the only non-trivial relative homology group of the pair $(D^n, S^{n-1})$ is in dimension $n$, where it is isomorphic to $\mathbb{Z}$. This tells us that the essential "relative content" of a solid disk is a sphere of the same dimension. This result is a cornerstone of the entire theory, forming a crucial step in an inductive method for calculating the homology of all spheres.

This method of collapsing is surprisingly versatile. Let's take a 2-sphere, $S^2$, and consider it relative to a subspace $A$ consisting of just two points—the north and south poles. What does this even mean? We are probing the structure of the sphere that connects these two points. Let's apply our trick: we collapse the subspace $A$ to a single point. Imagine pinching the north and south poles of a balloon together. What happens? The balloon is still a sphere, but now you've also created a loop, like a handle, running from the pinched point, through the balloon's interior, and back to the pinched point. The resulting space is a sphere and a circle joined at a point ($S^2 \vee S^1$). Relative homology brilliantly detects this new structure. The relative groups $H_n(S^2, A)$ are non-trivial in dimension 2 (detecting the sphere) and in dimension 1 (detecting the newly formed loop). We have used relative homology not just to see the original space, but to see the topological "pathways" between the points we chose to focus on.

This idea of isolating features is nowhere more powerful than in the construction of complex spaces from simple building blocks, a method known as building a CW complex. Imagine constructing a model with LEGO bricks. You start with a base, then add more bricks one by one. Cellular homology, which is built upon the foundation of relative homology, is the mathematical tool that allows us to track the contribution of each individual brick.

When we attach a new $k$-dimensional "cell" (a $k$-disk) to an existing space $X^{k-1}$ to form a new space $X^k$, the relative homology group $H_k(X^k, X^{k-1})$ precisely counts the number of $k$-cells we just added. For example, the familiar torus, $T^2$, can be built by starting with a point, attaching two 1-dimensional loops to get a figure-eight (which we can call the 1-skeleton, $A$), and then gluing on a single 2-dimensional patch. The relative homology group $H_2(T^2, A)$ turns out to be $\mathbb{Z}$, a single copy of the integers. That lone $\mathbb{Z}$ is the 2-cell we just glued on! Similarly, if we build the four-dimensional space $S^2 \times S^2$ by attaching a 4-cell to its lower-dimensional skeleton $S^2 \vee S^2$, the relative homology group $H_4(S^2 \times S^2, S^2 \vee S^2)$ is $\mathbb{Z}$, once again capturing the single top-dimensional cell we added.

This principle is completely general. It doesn't matter how complicated the attaching map is. If we form a space $X_n$ by attaching a 2-disk to a circle $S^1$ via a map that wraps the disk's boundary around the circle $n$ times, the relative homology $H_2(X_n, S^1)$ is still just $\mathbb{Z}$. The relative homology microscope focuses only on the cell itself, not on the contortions of its attachment. This makes it an incredibly effective tool for calculation, turning the daunting task of computing homology into a systematic, almost mechanical process of linear algebra. In a similar vein, the long exact sequence of a pair, a direct consequence of the definition of relative homology, provides an algebraic machine for computing groups. By feeding it the known homology groups of a space and a subspace, we can deduce the relative homology groups, as demonstrated in the calculation for the pair $(\mathbb{R}P^2, \mathbb{R}P^1)$.

In science, a null result can be as illuminating as a positive one. Finding that a quantity is zero tells you what isn't happening. So, what does it mean if all the relative homology groups of a pair $(X, A)$ are zero?

It means that, from the perspective of homology, the larger space $X$ contains no new topological features that weren't already present in the subspace $A$. This happens in a very specific situation: when $X$ is just a "thickened" version of $A$. More formally, if $A$ is a "deformation retract" of $X$, meaning you can continuously shrink $X$ onto $A$ without tearing it.

Consider a solid torus, $X=D^2 \times S^1$, which is like a donut with the hole filled in. Its "core" is a circle, $A$, running through its center. You can easily imagine shrinking the solid torus down onto this core circle. The inclusion of the circle into the solid torus is a homotopy equivalence. When we compute the relative homology groups $H_n(X,A)$, we find that they are all zero. The message is loud and clear: "Nothing new to see here!" The solid torus, despite being a 3-dimensional object containing a 1-dimensional circle, adds no new loops or voids that weren't, in a sense, already captured by the core circle. This sharpens our understanding immensely. Relative homology doesn't measure size or dimension in the usual sense; it measures the introduction of genuinely new topological structure.

The power of relative homology extends far beyond the internal affairs of pure topology. It provides a language to connect with other fields, such as knot theory, which has surprising ties to modern physics. A link, like the famous Hopf link, is a collection of circles embedded in our 3-dimensional space. A key question is how to distinguish one link from another.

We can use relative homology to probe the relationship between our ambient space, $\mathbb{R}^3$, and the link, $A$, that sits inside it. By analyzing the long exact sequence for the pair $(\mathbb{R}^3, A)$, we find that the relative homology groups are directly related to the absolute homology groups of the link itself. Since $\mathbb{R}^3$ is topologically simple (all its homology groups above dimension 0 are trivial), the sequence gives a clean result: $H_2(\mathbb{R}^3, A)$ is isomorphic to the group of 1-cycles of the link, and $H_1(\mathbb{R}^3, A)$ captures information about its connected components. For the Hopf link, a pair of interlinked circles, this computation yields algebraic invariants that help characterize its "linkedness". This is just a glimpse of how these abstract algebraic tools can be used to create concrete, computable invariants for physical objects like knotted DNA strands or complex molecules.

We have seen that relative homology is a powerful and versatile tool. But in science, we must always question our tools. How do we know our homology "machine" is reliable? What if we had built it in a different way? For instance, one can define homology using triangulations (simplicial homology) or using continuous maps (singular homology). It would be a disaster if these different definitions gave different answers for the same space.

Mathematicians, of course, worried about this too. They proved, with great effort, that for any single space, the different machines do agree. But what about our "relative" measurements? The proof that they also agree for any pair $(X, A)$ relies on a wonderfully elegant and powerful result from abstract algebra called the ​​Five Lemma​​. You can picture it as a logical device that governs two parallel assembly lines (the long exact sequences for simplicial and singular homology). The Five Lemma guarantees that if four out of five corresponding stations on the lines are synchronized, then the fifth station in the middle must also be synchronized.

This is more than just a technicality. It is our certificate of quality control. It ensures that the topological truths we uncover with our relative homology microscope are genuine features of the spaces themselves, not mere artifacts of the particular mathematical dialect we choose to speak. It reveals a deep and beautiful consistency at the very heart of mathematics, assuring us that the instrument we have so carefully constructed is indeed pointed at reality.