Tensor Products of Vector Bundles We begin by reviewing vector bundles. An n-dimensional vector bundle is a continuous map p (called the projection) from a space E (the total space) to the space B (the base space), p : E B , with a complex vector space structure n (where n is the fiber dimension) on p 1 b (the fiber over b) for each b B on which we have local trivializations . To define these local trivializations, we first cover B by open sets U I such that for each U , there exists a homeomorphism h : p 1 U U n p 1 b b n which is a vector space isomorphism for each b B . We can think of the local triviality condition as, in a small neighborhood of b B , the inverse image under p looks like the neighborhood with a copy of the vector space attached at each point. Intuitively, as each h must be continuous, if we take two points b and b’ in B that are close together, the fibers above b and b’ are also close together in E and any twisting that occurs going from the fiber p 1 b to the fiber p1 b ' is insignificant. If U i U j , we further require that the following diagram commutes: This induces a homeomorphism h j hi 1 : U i U j x, v where hi x : j h j 1 i n i n j n i U i U j x , h x (v ) n j j i is a homeomorphism on the fibers above x such that hij on each fiber. Here, we have distinguished between the fibers above x U i and the fibers above x U j , even though the fibers are isomorphic. The maps hi j are called transition functions and, intuitively, tell us how the fibers are glued together. Typically, we abuse terminology and refer to both the total space and the projection p as the “vector bundle”. Also, we often identify the base space with its image in the total space, don’t specify the trivializations, and don’t define the open cover of B. The majority of texts referenced did not mention the transition functions at all, let alone specifically. While discussing abuses of notation, we should mention the working idea behind vector bundles; we build them as follows. Take a copy of the base space. Attach a copy of the vector space to every point on the base space (this is our total space) in such a way that the movement in the total space from one point to another can be done smoothly. That means that the fibers must vary continuously over the base space. Finally, the map p is the projects the every point in the fiber over b back down onto b. Examples: Line bundle over S1 Let p1 : E1 S 1 where each fiber is 1 i two open sets U1 and U 2 . We want trivializations hi : p . We cover the circle with Ui Ui . Choose V0 and V1 disjoint open sets such that U 0 U1 V0 V1 . Define the gluing functions by h1 h01 : U0 U1 b, v U0 U1 b, v b, v b V1 b V0 Mobius bundle over S1 Let p1 : E1 S 1 where each fiber is . We cover the circle with two open sets U1 and U 2 . We want trivializations hi : pi1 Ui Ui . Choose V0 and V1 disjoint open sets such that U 0 U1 V0 V1 . Define the gluing functions by h1 h01 : U0 U1 b, v U0 U1 b, v b, v b V1 b V0 Note that the gluing functions define the difference between the bundles. Now that we have vector bundles, we begin constructing new bundles from old bundles. In class, we discussed the internal Whitney sum of vector bundles and subbundles. We can also form a sort of product of bundles, called the tensor product of vector bundles, which is important in K-theory. Due to the involved nature of the constructions of tensor products of vector bundles, we will look at three different versions. The tensor product to be the end result of any of these constructions. Construction 1: Let p1 : E1 B and p2 : E2 B be vector bundles over the same base space B . Consider the set defined by E1 E2 bB p11 b p21 b . For each open set U B over which E1 and E2 are trivial (that is, Ei U 1 i hi : p U U ni ni ), we choose isomorphisms . We define a topology U on p11 U p21 U by requiring that h1 h2 : p11 U p21 U U b p11 b p21 b n1 n1 n2 n2 be a homeomorphism on the fibers. Now, suppose that we have another pair of trivializations, ki : pi1 U U ni . Then hi pi1 U U where the isomorphism in the center is gi : U b, v ni U ni ni U ki pi1 U ni b, g b v i for each b U . Since the gi fix the first coordinate and map them as gi : U GLni ni to ni , we can view where the matrix might change over each fiber but changes continuously, fiber to fiber. Thus, composition with this invertible matrix takes hi to k i . Now, the entries of the matrices g1 b g2 b are the products of the entries of g1 b and g 2 b . We can compose in an analogous fashion: h b h b g b g b k b k b . 1 2 1 2 1 2 This means that the topology defined on p11 U p21 U by requiring that k1 k2 be a homeomorphism on the fibers is the same topology we get from requiring h1 h2 be a homeomorphism. The vector bundle constructed this way is the tensor product of vector bundles E1 and E2 over B, denoted E1 B E2 . This construction is a bit abstract for my taste, but does give the topology in a nice fashion. Construction 2: Now, we look at a second construction that is a bit more intuitive. In order to do so, we need to reconstruct the vector bundle as a quotient. Suppose we have vector a bundle p : E B and an open cover U I of B with local trivializations h : p1 U U n . We can think of E as the disjoint union of the images of the trivializations modulo an equivalence relation, E identifying U n with U n U n , where the relation vb is given by b, v vb h h1 b, v whenever b U U . Since the base point, b, in U is the same b in U , the mapping h h1 : U n U n is essentially a mapping from n to another copy of n . 1 Thus, we can think of h h as assigning a matrix to each point b U U that takes the first copy of n into the second copy of induced map g : U U GLn 1 1 . n . Thus, we identify h h1 with the Note that these maps satisfy a cocycle condition 1 g g h h h h h h g on U U U . These are the gluing functions for the vector bundle p : E B . Now, suppose we have p1 : E1 B (a vector bundle over B with fiber dimension n1 ) and p2 : E2 B (a vector bundle over B with fiber dimension n2 ) as above, save that we choose the open cover such that both bundles are trivial over the open cover. Then, we have gluing functions g i : U U GLni . We form the set E1 E2 bB 2 which p11 b p21 b as above and choose gluing functions g1 g 2 assigns to each b U U the tensor product of the matricies g1 b g b . i As g is induced by h h1 on each intersection, we have the local trivializations and the necessary structure for the vector bundle E1 E2 . Construction 3: The third construction is considerably more complicated, but a bit more elegant. We begin by defining the tensor product of two vector spaces. Let V be a vector space over with basis ei i 1 and W a vector space over n with basis f j m j 1 . Let we make the following identifications: V W ai , j vi w j | vi V , w j W i, j v v ' w v w v ' w v w w ' v w v w ' r v w rv w v rw for all v, v ' V , w, w ' W , and r . We call vi w j a simple tensor. The space V W is a vector space over with basis ei f j |1 i n, 1 j m . Now, suppose that V’ and W’ are also vector spaces. Let : V V ' and : W W ' be linear maps. We define : V W V ' W ' v w v w on simple tensors and extend linearly. Clearly, this is linear. Minimal work shows that it is well defined. We write the matrix for in terms of our ordered basis for V and call it A. Likewise, we write in terms of our ordered basis for W and call it B. Then, we can write A B in terms of the basis for V W by multiplying each entry in A by B. This gives an mn by mn matrix. For example, let b11 b12 b13 a11 a12 A B b21 b22 b23 a21 a22 b b 31 32 b33 Then, a11b11 a11b12 a11b13 a12b11 a12b12 a12b13 a11b21 a11b22 a11b23 a12b21 a12b22 a12b23 a B a12 B a11b31 a11b32 a11b33 a12b31 a12b32 a12b33 A B 11 a B a B 22 21 a21b11 a21b12 a21b13 a22b11 a22b12 a22b13 a b a b a21b23 a22b21 a22b22 a22b23 21 21 21 22 a b a b a b a b a b a b 21 32 21 33 22 31 22 32 22 33 21 31 Incidentally, this tensor product of matrices is sometimes called the Kronecker product. Next, we define a product more general than the tensor product and restrict to get the product we want. Let X and Y be compact topological spaces. Let A M n C X and B M m C Y . Then, we define the external tensor product to be the matrix in A B M nm C X Y defined by A B x, y A x B y for all x X , y Y . Now consider the vector bundles p1 : V X and p2 : W Y where X and Y are compact topological spaces. We write V Ran A and W Ran B for A an idempotent over X and B an idempotent over Y. Then, we define the external tensor product of V and W to be the vector bundle V W Ran A Ran B Ran A B over X Y . Note that if A and B are idempotents, so is A B (via some matrix multiplication) and that for each point x, y X Y , the fiber V W x, y Vx Wy . This is the tensor product of the vector spaces we discussed at the beginning of this construction. If we let X = Y for the vector spaces V and W, and let : X X X x x, x be the diagonal map, we can define the (internal) tensor product of V and W by V W * V W . Since the diagonal map is continuous, and since the pullback of a vector bundle by a continuous map is a vector bundle, the tensor product is also a vector bundle. After all of that, a picture or two might help with the intuition. A Jumping Off Point for K-theory: While the second construction is the most intuition (at least to me), the third leads us to the following very quickly: Corollary: Let X be a compact Hausdorff space. Then K 0 X is a commutative ring with unit under the internal tensor product. To prove this, we look at a more general proposition: Proposition: Let X and Y be compact Hausdorff spaces. The external tensor product determines a group homomorphism K 0 X K 0 Y K 0 X Y Pf: Given differences of equivalence classes of vector bundles V V ' K 0 X and W W ' K 0 Y , we define the product on simple tensors by V V ' W W ' V W V W ' V ' W V ' W ' and then extend by the distributive law to finite sums and constant multiples. We need to show that this product is well defined. To do so, we first suppose V V ' V V ' in K 0 X . Recall that V V ' V " V V 'V " for some vector bundle V " over X. Distributing the external tensor product across the internal Whitney sum and recalling the definition of E F , we have V W V ' W V " W V W V ' W V " W Thus, V W V ' W V W V ' W and V W V ' W V W V ' W . Hence, the product in K 0 X is well defined. We make the same argument to show that the product is well defined for representatives in K 0 Y . This definition also respects the defining relations for tensor products. The homomorphism is thus well defined. Now, we restrict to the case Y = X and look at the corollary. Corollary: Let X be a compact Hausdorff space. Then K 0 X is a commutative ring with unit under the internal tensor product. Pf: Letting Y = X in the proposition, we have vector bundles over the same space. From the third construction above, we know that the external tensor product becomes the internal tensor product when we pullback by the diagonal map. The tensor product of vector spaces is commutative and associative (see Dummit and Foote) and distributes across internal Whitney sums. The previous theorem and the proposition stated below imply that K 0 X is a commutative ring with multiplicative identity element 1 X , the trivial line bundle over X. The proposition to which we alluded in the proof of the corollary is this: Proposition: Let X and Y be compact Hausdorff spaces. Then the external tensor product defines a map from Idem C X Idem C Y to Idem C X Y . For more information about any of these constructions, or if there are errors and you would like to see a correct version, please see Dummit, David and Foote, Richard. Abstract Algebra. Prentice Hall Publishing Park, Efton. Complex Topological K-Theory. Cambridge UP. Print. Hatcher, Allen. Vector Bundles and K-Theory. Copyright Allen Hatcher Osborn, Howard. Vector Bundles: Foundations and Stieffel-Whitney Classes. Academic Press, Inc. Print. ö