Orientations of Vector Spaces

Consistent Orientation of Bases Orientation of a Vector Space Orientation Through a Top-Degree Covector

Consistent Orientation of Bases

Everyone who has drawn coordinate axes has already made a choice of orientation without naming it. A number line is drawn with the positive direction to the right; a pair of plane axes is drawn so that the rotation carrying the first toward the second runs counterclockwise; the three axes of space are drawn by the right-hand rule, the thumb pointing along the third when the fingers curl from the first to the second. Each rule singles out a preferred family of ordered bases and rejects its mirror image. The words "to the right," "counterclockwise," and "right-handed" are not mathematics, but they all encode the same algebraic fact: the preferred bases are exactly those whose change-of-basis matrix from the standard one has positive determinant.

In an abstract vector space there is no standard basis to measure against, and no canonical sense of "right-handed." Among the ordered bases of the space of real polynomials of degree at most two, for instance, neither \((1, x, x^2)\) nor \((x^2, x, 1)\) has any prior claim to being correctly oriented. What survives the loss of a reference basis is not the notion of a single basis being correct, but the relation of two bases agreeing with each other. That relation is the foundation on which everything in this page rests, and it is defined by a determinant of sign.

Definition: Consistently Oriented Bases

Let \(V\) be a real vector space of dimension \(n \geq 1\), and let \((E_1, \dots, E_n)\) and \((\widetilde{E}_1, \dots, \widetilde{E}_n)\) be two ordered bases of \(V\). The transition matrix \(\bigl(B_i^{\,j}\bigr)\) from the second basis to the first is the unique array of real numbers determined by \[ E_i = B_i^{\,j}\, \widetilde{E}_j , \] expressing each old basis vector in terms of the new one (summation over the repeated index \(j\) is understood). The two bases are consistently oriented if \(\det\bigl(B_i^{\,j}\bigr) > 0\).

The transition matrix is always invertible — it carries one basis to another — so its determinant is nonzero, and the only question is its sign. Reversing the roles of the two bases replaces the matrix by its inverse, whose determinant has the same sign, so the relation does not depend on which basis is named first. These observations are the substance of the following fact.

Proposition: Two Orientation Classes

Consistent orientation is an equivalence relation on the set of all ordered bases of an \(n\)-dimensional real vector space \(V\), and it has exactly two equivalence classes.

Proof.

Reflexivity holds because the transition matrix from a basis to itself is the identity, whose determinant is \(1 > 0\). For symmetry, if \((E_i)\) and \((\widetilde{E}_i)\) have transition matrix \(B\), then \((\widetilde{E}_i)\) and \((E_i)\) have transition matrix \(B^{-1}\), and \(\det(B^{-1}) = (\det B)^{-1}\) by the multiplicativity of the determinant; a positive number has a positive reciprocal. For transitivity, suppose a third basis passes to the second through a transition matrix \(C\), and the second passes to the first through \(B\). Substituting one change of basis into the other shows that the third basis passes directly to the first through the product \(BC\); the transition matrices compose by matrix multiplication. Their determinants therefore multiply, again by the multiplicativity of the determinant, and a product of two positive numbers is positive.

To count the classes, fix one basis \((E_i)\). Any basis \((\widetilde{E}_i)\) has a transition matrix to \((E_i)\) of nonzero determinant, hence of determinant either positive or negative; these two cases are precisely whether the basis is consistently oriented with \((E_i)\) or not. Both cases occur: the basis \((E_i)\) itself sits in the positive case, while replacing \(E_1\) by \(-E_1\) multiplies the determinant by \(-1\) and produces a basis in the negative case. Thus there are exactly two classes — the bases consistently oriented with \((E_i)\), and those consistently oriented with \((-E_1, E_2, \dots, E_n)\).

Orientation of a Vector Space

With the bases sorted into two classes, an orientation is simply the act of choosing one of them. Nothing distinguishes the two classes intrinsically — the choice is a piece of extra data attached to the space, exactly as the labels "right-handed" and "left-handed" are conventions imposed on physical space rather than features read off from it. Once the choice is made, every basis acquires a sign.

Definition: Orientation of a Vector Space

Let \(V\) be a real vector space of dimension \(n \geq 1\). An orientation of \(V\) is a choice of one of the two consistent-orientation classes of ordered bases. An ordered basis \((E_1, \dots, E_n)\) determines the orientation given by its own class, written \([E_1, \dots, E_n]\); the other orientation is written \(-[E_1, \dots, E_n]\). A vector space together with a choice of orientation is an oriented vector space. Once an orientation is fixed, a basis lying in the chosen class is called positively oriented (or simply oriented), and a basis in the other class is negatively oriented.

The definition presumes \(n \geq 1\), since a positive-dimensional space is the only kind with ordered bases to sort. The zero-dimensional case has a single basis — the empty one — and so cannot be handled by counting basis classes; it is given its own convention, chosen now so that later statements about boundaries and determinants extend without exception to dimension zero. An orientation of a zero-dimensional vector space is a choice of one of the numbers \(+1\) or \(-1\).

Definition: The Standard Orientation of \(\mathbb{R}^n\)

The standard orientation of \(\mathbb{R}^n\) for \(n \geq 1\) is the orientation \([e_1, \dots, e_n]\) determined by the standard basis. Under the usual conventions for drawing axes, the positively oriented bases are exactly the familiar ones: for \(\mathbb{R}^1\) a basis pointing in the positive direction, for \(\mathbb{R}^2\) a basis whose rotation from the first vector to the second is counterclockwise, and for \(\mathbb{R}^3\) a right-handed basis. The standard orientation of \(\mathbb{R}^0\) is defined to be \(+1\).

These three pictures are not independent stipulations but consequences of the single algebraic rule: in each case a basis is positively oriented precisely when its transition matrix from the standard basis has positive determinant. The everyday words "rightward," "counterclockwise," and "right-handed" may therefore be taken as their rigorous definitions, valid in the three dimensions where intuition reaches and superseded by the determinant in every dimension where it does not.

Orientation Through a Top-Degree Covector

An orientation as defined so far is a combinatorial object: a choice between two classes of bases. There is a second, analytic description that will prove far more useful once vector spaces give way to manifolds, where bases vary from point to point but a single differential form can be specified globally. The bridge is the space of top-degree alternating tensors \(\Lambda^n(V^*)\), which on an \(n\)-dimensional space is one-dimensional: removing the origin leaves two connected rays, and choosing a ray turns out to be the same as choosing an orientation.

Proposition: Orientation Determined by a Top-Degree Covector

Let \(V\) be a real vector space of dimension \(n\). Each nonzero \(\omega \in \Lambda^n(V^*)\) determines an orientation \(\mathcal{O}_\omega\) of \(V\) as follows. If \(n \geq 1\), then \(\mathcal{O}_\omega\) is the set of ordered bases \((E_1, \dots, E_n)\) with \[ \omega(E_1, \dots, E_n) > 0 . \] If \(n = 0\), then \(\Lambda^0(V^*) = \mathbb{R}\), and \(\mathcal{O}_\omega\) is \(+1\) when \(\omega > 0\) and \(-1\) when \(\omega < 0\). Two nonzero top-degree covectors determine the same orientation if and only if each is a positive multiple of the other.

Proof.

The zero-dimensional case is immediate, since a nonzero element of \(\Lambda^0(V^*) = \mathbb{R}\) is just a nonzero real number, whose sign is the orientation. Assume \(n \geq 1\), and let \(\omega\) be a nonzero element of \(\Lambda^n(V^*)\). Because \(\Lambda^n(V^*)\) is one-dimensional, \(\omega\) does not vanish on any basis, so for every ordered basis the number \(\omega(E_1, \dots, E_n)\) is nonzero, and the sign condition \(\omega(E_1, \dots, E_n) > 0\) partitions the bases into two sets. It remains to show that the set \(\mathcal{O}_\omega\) on which \(\omega\) is positive is exactly one consistent-orientation class.

Let \((E_i)\) and \((\widetilde{E}_i)\) be two ordered bases, and let \(T : V \to V\) be the linear map sending \(E_i\) to \(\widetilde{E}_i\). Then \(T\) is the change-of-basis map, and by the top-degree alternation–determinant identity, \[ \omega(\widetilde{E}_1, \dots, \widetilde{E}_n) = \omega(TE_1, \dots, TE_n) = (\det T)\, \omega(E_1, \dots, E_n) . \] The determinant of \(T\) is nonzero, since \(T\) carries a basis to a basis, and its sign is exactly what records consistent orientation: the two bases are consistently oriented if and only if \(\det T > 0\). (The transition matrix \(\bigl(B_i^{\,j}\bigr)\) of the earlier definition is the matrix of \(T^{-1}\), whose determinant has the same sign as \(\det T\), so the two formulations of consistent orientation agree.) By the displayed identity, \(\det T > 0\) holds if and only if \(\omega(\widetilde{E}_1, \dots, \widetilde{E}_n)\) and \(\omega(E_1, \dots, E_n)\) have the same sign. Hence the bases on which \(\omega\) is positive form a single consistent-orientation class, and \(\mathcal{O}_\omega\) is an orientation.

Finally, let \(\omega'\) be any other nonzero element of \(\Lambda^n(V^*)\). Since this space is one-dimensional, \(\omega' = c\,\omega\) for some nonzero scalar \(c\). For any basis, \(\omega'(E_1, \dots, E_n) = c\,\omega(E_1, \dots, E_n)\), so \(\omega'\) and \(\omega\) are positive on the same bases exactly when \(c > 0\). Thus \(\mathcal{O}_{\omega'} = \mathcal{O}_\omega\) if and only if \(\omega'\) is a positive multiple of \(\omega\).

The proposition closes the loop between the combinatorial and analytic pictures: an orientation is a choice of basis class, equivalently a choice of one of the two rays in \(\Lambda^n(V^*) \setminus \{0\}\). A nonzero top-degree covector that determines a given orientation in this way is called a positively oriented \(n\)-covector for that orientation; on \(\mathbb{R}^n\) with the standard orientation, the covector \(e^1 \wedge \dots \wedge e^n\) is positively oriented. The analytic description is the one that survives the passage to manifolds, where a nowhere-vanishing top-degree form will pin down an orientation at every point at once — the device that makes integration on a manifold possible.