Loops on the Circle and Their Lifts
As a loop travels around the circle, it may go around once, several times, or travel forward and then back. The endpoint of the loop alone cannot distinguish these possibilities: every based loop starts and ends at the same point. A lift of the loop to the real line records the travel before it is wrapped onto the circle. Its endpoint will provide an integer that counts the loop’s net winding.
Let \(S^1=\{z\in\mathbb{C}:|z|=1\}\) have the subspace topology inherited from the usual topology on \(\mathbb{C}\), and use \(1\in S^1\) as basepoint. We use the covering map \(p:\mathbb{R}\to S^1\) given by \(p(t)=e^{2\pi i t}\), where \(\mathbb{R}\) has its usual topology. In particular, \(p(t)=1\) exactly when \(t\in\mathbb{Z}\).
The standard path-lifting result for the covering map \(p\) says that every path \(\alpha:I\to S^1\) with \(\alpha(0)=1\) has a unique continuous lift \(\widetilde{\alpha}:I\to\mathbb{R}\) satisfying \(\widetilde{\alpha}(0)=0\) and \(p\circ\widetilde{\alpha}=\alpha\). We call this the normalized lift. If \(\alpha\) is a based loop, then \(p(\widetilde{\alpha}(1))=\alpha(1)=1\); hence \(\widetilde{\alpha}(1)\in\mathbb{Z}\).
Winding Number and the Loop Class
To assign a winding number to a class of loops, rather than to a chosen representative, we need the number to be unchanged under based homotopy. The homotopy invariance of winding number is established in the earlier material on homotopy invariance of winding number: if two based loops are homotopic through based loops, their normalized lifts have the same endpoint. Thus \(w([\alpha])=w(\alpha)\) is well defined on \(\pi_1(S^1,1)\).
Worked Example: A Loop That Goes Around Three Times
Consider \(\alpha(t)=e^{6\pi i t}\) for \(t\in I\). It is a based loop because \(\alpha(0)=e^0=1\) and \(\alpha(1)=e^{6\pi i}=1\). The function \(\widetilde{\alpha}(t)=3t\) starts at \(0\), and \(p(\widetilde{\alpha}(t))=e^{2\pi i(3t)}=e^{6\pi i t}=\alpha(t)\). By uniqueness, this is the normalized lift. Therefore \(w(\alpha)=\widetilde{\alpha}(1)=3\).
Check: the lift endpoint is \(3\), and \(p(3)=e^{6\pi i}=1=\alpha(1)\), as required for a based loop.
Worked Example: A Piecewise Loop with Backtracking
Define a continuous function \(\theta:I\to\mathbb{R}\) by \(\theta(t)=\frac{9}{2}t\) for \(0\le t\le\frac13\), \(\theta(t)=\frac32-6(t-\frac13)\) for \(\frac13\le t\le\frac23\), and \(\theta(t)=-\frac12+\frac{15}{2}(t-\frac23)\) for \(\frac23\le t\le1\). At \(t=\frac13\), the first formula gives \(\frac32\) and the second gives \(\frac32-6(0)=\frac32\). At \(t=\frac23\), the second gives \(\frac32-6(\frac13)=-\frac12\) and the third gives \(-\frac12+\frac{15}{2}(0)=-\frac12\). Thus the pieces join continuously. The last formula gives \(\theta(1)=-\frac12+\frac{15}{2}(\frac13)=-\frac12+\frac52=2\).
Set \(\alpha(t)=p(\theta(t))\). Since \(\theta(0)=0\) and \(\theta(1)=2\), we have \(\alpha(0)=p(0)=1\) and \(\alpha(1)=p(2)=1\). The function \(\theta\) is the normalized lift, so \(w(\alpha)=2\). Although the lift moves from \(0\) to \(\frac32\), then back to \(-\frac12\), its net change is \(2\).
Check: applying the covering map to the lift’s endpoint gives \(p(\theta(1))=p(2)=e^{4\pi i}=1=\alpha(1)\).
Worked Example: A Loop That Goes Forward and Back
Let \(\theta(t)=2t\) on \(0\le t\le\frac12\) and \(\theta(t)=2-2t\) on \(\frac12\le t\le1\), and set \(\alpha(t)=p(\theta(t))\). The two formulas agree at \(t=\frac12\): both give \(1\). Also, \(\theta(0)=0\), \(\theta(1)=0\), so \(\alpha\) is a based loop and its normalized lift is \(\theta\). Its winding number is \(w(\alpha)=0\).
In this case the loop can be contracted explicitly through based loops. Define \(H(s,t)=p((1-s)\theta(t))\). This is continuous, and \(H(0,t)=\alpha(t)\). At \(s=1\), \(H(1,t)=p(0)=1\) for every \(t\), the constant loop. For every \(s\), both endpoint values are \(p((1-s)0)=1\), so \(H\) fixes the basepoint. This example illustrates that a loop may move around the circle and still represent the identity class if its net winding is zero.
Check: at \(t=1\), the lift is \(\theta(1)=2-2(1)=0\), and \(p(0)=1=\alpha(1)\).
Concatenation Adds Winding Numbers
For based loops \(\alpha,\beta\), their concatenation \(\alpha*\beta\) is defined by \((\alpha*\beta)(t)=\alpha(2t)\) for \(0\le t\le\frac12\) and \((\alpha*\beta)(t)=\beta(2t-1)\) for \(\frac12\le t\le1\). The pieces agree at the joining time because \(\alpha(1)=1=\beta(0)\).
Proof. Let \(A\) and \(B\) be the normalized lifts of \(\alpha\) and \(\beta\). Write \(m=A(1)\) and \(n=B(1)\), both integers. Define \(L:I\to\mathbb{R}\) by \(L(t)=A(2t)\) for \(0\le t\le\frac12\) and \(L(t)=m+B(2t-1)\) for \(\frac12\le t\le1\). At \(t=\frac12\), the first formula gives \(A(1)=m\), and the second gives \(m+B(0)=m\), so \(L\) is continuous. It starts at \(A(0)=0\). On the first half, \(p(L(t))=\alpha(2t)\). On the second half, \(p(L(t))=p(m+B(2t-1))=p(B(2t-1))=\beta(2t-1)\), since \(m\in\mathbb{Z}\) and \(p(u+m)=p(u)\). Thus \(L\) is the normalized lift of \(\alpha*\beta\), and \(L(1)=m+B(1)=m+n\). This proves the winding-number formula. Since \(W\) is well defined on classes and concatenation represents the group operation, the formula says exactly that \(W\) is a homomorphism. \(\square\)
Worked Example: Concatenating Loops with Winding Numbers 2 and −1
Let \(\alpha(t)=p(2t)\) and \(\beta(t)=p(-t)\). Their normalized lifts are \(A(t)=2t\) and \(B(t)=-t\), so \(w(\alpha)=2\) and \(w(\beta)=-1\). The lift of the concatenation is \(L(t)=4t\) for \(0\le t\le\frac12\), and \(L(t)=2-(2t-1)=3-2t\) for \(\frac12\le t\le1\). Both pieces equal \(2\) at \(t=\frac12\). The endpoint is \(L(1)=3-2=1\), so \(w(\alpha*\beta)=1=2+(-1)\).
Check: \(p(L(1))=p(1)=e^{2\pi i}=1\), so the lift endpoint gives the correct endpoint of the concatenated based loop.
The Fundamental Group of the Circle
Proof. The map is a homomorphism by the concatenation result. We prove that it is surjective and injective.
For surjectivity, let \(n\in\mathbb{Z}\) be arbitrary, including \(n=0\) or a negative integer. Define \(\gamma_n(t)=p(nt)\). Then \(\gamma_n(0)=p(0)=1\) and \(\gamma_n(1)=p(n)=1\), so it is a based loop. Its normalized lift is \(t\mapsto nt\), whose endpoint is \(n\). Therefore \(W([\gamma_n])=n\), and every integer occurs.
For injectivity, suppose \(W([\alpha])=W([\beta])=n\). Let \(A,B:I\to\mathbb{R}\) be the normalized lifts. They have the same endpoints: \(A(0)=B(0)=0\) and \(A(1)=B(1)=n\). Define \(H(s,t)=p((1-s)A(t)+sB(t))\). This is continuous. For every \(s\), its value at \(t=0\) is \(p(0)=1\), and its value at \(t=1\) is \(p((1-s)n+sn)=p(n)=1\). Also, \(H(0,t)=\alpha(t)\) and \(H(1,t)=\beta(t)\). Thus \(H\) is a homotopy through based loops, so \([\alpha]=[\beta]\). The map \(W\) is injective and therefore is an isomorphism. \(\square\)
This result computes the group, not just the winding number of individual examples: two based loops represent the same element of \(\pi_1(S^1,1)\) exactly when they have the same winding number. In particular, the constant loop corresponds to \(0\), and concatenation of classes corresponds to addition in \(\mathbb{Z}\).
A Common Pitfall
Common mistake: Since a based loop ends at \(1\), one might conclude that the endpoint of its lift must be \(0\). That confuses the endpoint in \(S^1\) with the endpoint in \(\mathbb{R}\). For example, the loop \(\alpha(t)=p(2t)\) ends at \(p(2)=e^{4\pi i}=1\), but its normalized lift ends at \(2\), not \(0\). The integer is essential: reducing the lifted endpoint modulo integers erases the information that distinguishes different loop classes.
Check Your Understanding
Use normalized lifts to answer these questions.
- What is the winding number of the loop \(\alpha(t)=e^{-8\pi i t}\), and what is its normalized lift?
- Give a based loop with winding number \(4\), and verify its lift endpoint.
- If two based loops have normalized lifts with the same endpoint, why are the loops homotopic through based loops?
- For loops with winding numbers \(-2\) and \(5\), what is the winding number of their concatenation?
- Why does knowing only that a loop ends at \(1\) not determine its winding number?