HAMILTONIAN PATHS IN PROJECTIVE CHECKERBOARDS Margaret H. Forbush, Elizabeth Hanson, Susan Kim, Andrew Mauer-Oats, Rhian Merris, Jennifer Oats-Sargent

Transcription

1 HAMILTONIAN PATHS IN PROJECTIVE CHECKERBOARDS Margaret H. Forbush, Elizabeth Hanson, Susan Kim, Andrew Mauer-Oats, Rhian Merris, Jennifer Oats-Sargent, Seth Oldham, Kate Sharkey, and Dave Witte Abstract. Place a checker in some square of an nn checkerboard. The checker is allowed to step either to the east or to the north, and is allowed to step o the edge of the board in a manner suggested by the usual identication of the edges of the square to form a projective plane. We give an explicit description of all the routes that can be taken by the checker to visit each square exactly once. 1. Introduction Place a checker in some square of an n n checkerboard. The checker is allowed to step (a distance of one square) either to the east or to the north in particular, it does not move diagonally. We ask whether it is possible to move the checker in this manner through a route that visits each square exactly once. We feel free to adopt graph-theoretic terminology whenever it is convenient, so we refer to such a route as a hamiltonian path. It is traditional to allow the checker to step o the edge of the board (otherwise, it is clear that no hamiltonian path exists). Namely, when the checker steps o the east edge of the board, it moves to the westernmost square of the same row; when the checker steps o the north edge of the board, it moves to the southernmost square of the same column. This can be described succinctly by saying that the checkerboard has been made into a torus, by gluing each of its edges to the opposite edge. For these toroidal checkerboards, it is not dicult to see that every square is the starting point of some hamiltonian path, and several authors have studied more delicate properties of paths and cycles in toroidal checkerboards [1], [2], [4], [5], [6], [7]. In this paper, we do not study toroidal checkerboards. Instead, the checker is allowed to step o the edge of the board in a somewhat dierent manner, corresponding to the usual procedure for creating a projective plane by applying a twist when gluing each edge of a square to the opposite edge Denition. The squares of the nn checkerboard can be naturally identied with the set B n of ordered pairs (p; q) of integers with 0 p; q n? 1. Dene E: B n! B n and N: B n! B n by (p; q)e = (p + 1; q) if p < n? 1 (0; n? 1? q) if p = n? 1 Corresponding author: D. Witte. Received September Revised December 1996, August 1997, and February Typeset by AMS-TEX

2 2 MARGARET H. FORBUSH ET AL. (0; 8) (8; 8) (0; 0) (8; 0) (0; 8) (8; 8) (0; 0) (8; 0) and Figure 1. Locations of the initial () and terminal ( ) squares of hamiltonian paths on the 9 9 projective checkerboard. (p; q)n = (p; q + 1) if q < n? 1 (n? 1? p; 0) if q = n? 1 The n n projective checkerboard B n is the digraph whose vertex set is B n, with a directed edge from to E and from to N, for each 2 B n. We usually refer to the vertices of B n as squares. For these projective checkerboards, we show that only certain squares can be the initial square of a hamiltonian path, and only certain squares can be the terminal square. The form of the answer depends on the parity of n; illustrative examples appear in Figures 1 and Notation. For convenience, we let 6n = bn=2c and 6n? = b(n? 1)=2c. Thus, if n is odd, we have 6n = 6n? = (n? 1)=2, whereas, if n is even, we have 6n = n=2 and 6n? = 6n? Theorem. A square (p; q) of B n (with 0 p; q n? 1) is the initial square of a hamiltonian path if and only if either (1) q = 0 and p 6n? ; or (2) q = 6n? and p 6n; or (3) q = 6n and p 6n? ; or (4) p = 0 and q 6n? ; or (5) p = 6n? and q 6n; or (6) p = 6n and q 6n?. It is the terminal square of a hamiltonian path if and only if either (1) q = n? 1 and p 6n; or (2) q = 6n and p 6n? ; or (3) q = 6n? and p 6n; or (4) p = n? 1 and q 6n; or (5) p = 6n and q 6n? ; or (6) p = 6n? and q 6n. Furthermore, we give an explicit description of all the hamiltonian paths in B n. This more detailed result is stated in Section 2, after the necessary terminology has been introduced. The proof is in Section 3.

3 HAMILTONIAN PATHS IN PROJECTIVE CHECKERBOARDS 3 (0; 9) (9; 9) (0; 0) (9; 0) (0; 9) (9; 9) (0; 0) (9; 0) Figure 2. Locations of the initial () and terminal ( ) squares of hamiltonian paths on the projective checkerboard. 2. Direction-forcing diagonals 2.1. Denition. Dene a symmetric, reexive relation on the set of squares by if fe; Ng \ fe; Ng 6= ;: The equivalence classes of the transitive closure of are direction-forcing diagonals. For short, we refer to them simply as diagonals. Thus, the diagonal containing is f; NE?1 ; (NE?1 ) 2 ; : : : ; EN?1 g: 2.2. Notation. For 2 B n, we let x and y be the components of, that is, = ( x ; y ) Notation. For 0 i 2n? 2, let S i = f 2 B n j x + y = ig Proposition. For each i with 0 i 2n? 3, the set D i = S i [ S 2n?3?i is a diagonal. The only other diagonal D 2n?2 consists of the single square (n? 1; n? 1). Notice that D i = D 2n?3?i if 0 i 2n? Notation. Let D be a diagonal, other than D 2n?2. Then, from Proposition 2.4, we may write D = S a [ S b with a < b. We let D? = S a and D + = S b. To nd the hamiltonian paths in B n, it is helpful to focus attention on the following class of subgraphs of B n, which contains all the hamiltonian paths Denition. A spanning quasi-path is a spanning subdigraph H of B n, such that exactly one component of H is a (directed) path, and each of the other components of H is a (directed) cycle. We usually use to denote the initial square of the path, and to denote the terminal square. Thus, a spanning quasi-path has two distinguished vertices and, such that every square except has indegree 1, and every square except has outdegree 1. The square has indegree 0, and the square has outdegree 0.

4 4 MARGARET H. FORBUSH ET AL Denition. If H is a spanning quasi-path, then the diagonal containing the terminal square is called the terminal diagonal of H. All other diagonals are non-terminal diagonals. The following fundamental propositions, which are used throughout the paper, justify the name direction-forcing diagonal, and reveal the special status of the terminal diagonal. The results are adapted from ideas of Housman [3] Denition. Let H be a spanning quasi-path in B n. A square travels east (in H) if the edge from to E is in H. Similarly, travels north (in H) if the edge from to N is in H Proposition (cf. [1, Lemma 6.4c]). If H is a spanning quasi-path in B n, then, for each non-terminal diagonal D, either every square in D travels north, or every square in D travels east. For short, we say that either D travels north or D travels east Proposition (cf. [1, Lemma 6.4b]). Let T be the terminal diagonal of a spanning quasi-path H in B n, with initial vertex and terminal vertex, and let 2 T. (1) if N 6=, then NE?1 travels east; (2) if E 6=, then EN?1 travels north; (3) if travels east and N 6=, then NE?1 travels east; (4) if travels east, then EN?1 does not travel north; (5) if travels north and E 6=, then EN?1 travels north; and (6) if travels north, then NE?1 does not travel east Corollary (cf. [1, Lemma 6.4a]). If H is a spanning quasi-path in B n, then the diagonal that contains E?1 and N?1 is the terminal diagonal. The following corollary follows from Proposition 2.10 by induction Corollary. Let T be the terminal diagonal of a spanning quasi-path H in B n, with initial vertex and terminal vertex, and let jt j denote the cardinality of T. For each 2 T, there is a unique integer u() 2 f1; 2; : : : ; jt jg with = (NE?1 ) u() ; the square travels east if and only if u() < u(e?1 ). (Similarly, there is a unique integer v() 2 f1; 2; : : : ; jt jg with = E?1 (EN?1 ) v() ; the square travels east if and only if v() < v().) From Corollary 2.12, we see that the location of the initial square and terminal square determine which vertices in the terminal diagonal travel east, and which of them travel north, so we have the following corollary Corollary. A hamiltonian path (or, more generally, a spanning quasi-path) is uniquely determined by specifying (1) its initial vertex; (2) its terminal vertex; and (3) which of its non-terminal diagonals travel east. Our main results determine precisely which choices of (1), (2), and (3) lead to a hamiltonian path, rather than some other spanning quasi-path. The following theorem shows that, to nd the possible initial squares and terminal squares of hamiltonian paths, one need only consider hamiltonian paths of two special types: those in which all non-terminal diagonals travel east, and those in which all non-terminal diagonals travel north.

5 HAMILTONIAN PATHS IN PROJECTIVE CHECKERBOARDS 5 initial vertex terminal vertex path directions (n? k; 0) 1 k 6n (n? k; 0) 1 k 6n + 1 (6n? k; 6n) (E 2n?1 N) 6n E n?1 (6n + k? 1; 6n) (E 2n?1 N) k?1 E? 2k?2 NE 2i?1 NE 2(n?k?i+1) NE 2k?2 6n?k+1 i=1 (0; n? 1) (6n; 6n) (E 2n?1 N) 6n E n?1 (0; n? 1) (6n; 6n? 1)? N 2 E 2i NE 2(n?i)?1 6n i=1 Figure 3. Some hamiltonian paths in B n, when n is odd (where 6n = (n? 1)=2). All non-terminal diagonals travel east. initial vertex terminal vertex path directions (n? k; 0) 1 k 6n (6n? k; 6n? 1) (E2n?1 N) 6n?1 E 2n?1 (n? k; 0) 1 k 6n (6n + k? 2; 6n) (E 2n?1 N) k?1 E? 2k?2 NE 2i?1 NE 2(n?k?i+1) NE 2k?2 6n?k NE n?2k+1 NE 2k?2 (6n? 1; 0) (n? 1; 6n) (E 2n?1 N) 6n?1 E 2n?1 (0; n? 1) (6n? 1; 6n? 1)? N 2 E 2i NE 2n?2i?1 6n?1 i=1 N 2 E n?1 (0; n? 1) (6n; 6n? 1) (E 2n?1 N) 6n?1 E 2n?1 i=1 Figure 4. Some hamiltonian paths in B n, when n is even (where 6n = n=2). All non-terminal diagonals travel east Theorem. If there is a hamiltonian path from to, then there is a hamiltonian path from to, such that either all non-terminal diagonals travel east, or all non-terminal diagonals travel north. Figures 3 and 4 list several hamiltonian paths. We now dene some simple transformations that can be applied to create additional hamiltonian paths from these Denition. For = ( x ; y ) 2 B n, let ~ = (n? 1? x ; n? 1? y ) be the inverse of, and let = ( y ; x ) be the transpose of Proposition. Let [ 0 ; : : : ; m ] be the sequence of squares visited by a hamiltonian path H in B n, from = 0 to = m. Then the following are also hamiltonian paths in B n : the inverse of H from ~ to ~: [~ m ; ~ m?1 ; : : : ; ~ 0 ]; the transpose of H from to : [ 0 ; : : : ; m];

6 6 MARGARET H. FORBUSH ET AL. the inverse-transpose of H from ~ to ~ : [~ m ; ~ m?1 ; : : : ; ~ 0]. Our main theorem can be stated as follows Main Theorem. If H is any hamiltonian path in B n, such that all nonterminal diagonals travel east, then either H is one of the hamiltonian paths listed in Figures 3 and 4, or H is the inverse of one of these hamiltonian paths. By considering the transposes of these paths, we obtain the following corollary Corollary. If H is any hamiltonian path in B n, such that all non-terminal diagonals travel north, then H is either the transpose or the inverse-transpose of one of the hamiltonian paths listed in Figures 3 and 4. The Main Theorem and its corollary, combined with Theorem 3.6 0, suce to determine all the initial and terminal squares of hamiltonian paths. These conclusions are detailed in Theorem 1.3. By combining the Main Theorem with Proposition 3.3 and Theorem 3.5 below, we obtain the following precise description of the hamiltonian paths in B n, not just those in which all non-terminal diagonals travel the same direction. The point is that all hamiltonian paths can be derived by applying certain simple transformations to the basic paths listed in Figures 3 and Corollary. The following construction always results in a hamiltonian path in B n, and every hamiltonian path can be constructed in this manner. (1) Start with a basic hamiltonian path H 1 from Figure 3 or 4. (2) Write the terminal diagonal of H 1 in the form S a [ S b (with a < b), and let N be any subset of fd j j 0 j < ag. (3) Let H 2 be the spanning quasi-path of B n in which the non-terminal diagonals in N travel north, all the other non-terminal diagonals travel east, and each square in the terminal diagonal travels exactly the same way as it does in H 1. (4) Let H be H 2, or the inverse, or transpose, or inverse-transpose of H 2. Then H is a hamiltonian path in B n.

7 HAMILTONIAN PATHS IN PROJECTIVE CHECKERBOARDS 7 3. The Main Theorem In this section, we prove Main Theorem Notation. We use [](X 1 X 2 : : : X m ), where X i 2 fe; Ng, to denote the path (or cycle) in B n that visits the squares ; X 1 ; X 1 X 2 ; : : : ; X 1 X 2 : : : X m : Note that a 2 2 projective checkerboard has hamiltonian cycles, namely, the cycles (NNNN) and (EEEE). However, the following theorem shows that this is the only such board; for n > 2, there is no hamiltonian cycle in B n Theorem. On an n n projective checkerboard with n 3, there are no hamiltonian paths with initial square = (0; 0). Therefore, D 2n?2 is not the terminal diagonal of any hamiltonian path in B n. Proof. Suppose H is a hamiltonian path in B n, with initial square = (0; 0). From Corollary 2.11, we see that the terminal square must be (n? 1; n? 1). Let [ 0 ; 1 ; : : : ; m ] be the sequence of squares visited by H. By induction on i, we show that m?i = ~ i (the inverse of i ) for 0 i m. Assume inductively that m?i = ~ i. Suppose, without loss of generality, that i travels east in H. From Proposition 2.4, it is not dicult to see, for all 2 B n, that the squares and ~E?1 are in the same diagonal. Therefore, ~ i E?1 must travel east, as i does. Then, since (~ i E?1 )E = ~ i = m?i, we conclude that ~ i E?1 = m?i?1. Therefore, m?i?1 = ~ i E?1 = g i E = ~ i+1 ; as desired. It is clear that H steps o the edge of the board somewhere, for, otherwise, it could not visit more than 2n? 1 squares. Let (p; q) be the square from which H steps o the edge of the board (the rst time). We may assume, without loss of generality, that p = n? 1, and that H steps east, from n?1+q = (n? 1; q) to n+q = (0; n? 1? q). Then, from the conclusion of the preceding paragraph, we have m?(n?1+q) = ~ n?1+q = (n? 1; q) = (0; n? 1? q) = n+q : Therefore, m? (n? 1 + q) = n + q. Since (p; q) 6= = (n? 1; n? 1) and p = n? 1, we must have q n? 2, so this implies that m = 2n? 1 + 2q 4n? 5. Because n 3, we have 4n? 5 < n 2? 1, so this contradicts the fact that the length of a hamiltonian path must be n 2? 1. Henceforth, we may (and do) always assume that the terminal diagonal T of a hamiltonian path H is not D 2n?2. Therefore, Proposition 2.4 implies that we may write T in the form T = S a [ S b, with a < b Proposition. Suppose the terminal diagonal of a hamiltonian path H is S a [ S b, with a < b. Then all D j, where a < j < b, travel in the same direction. Proof by induction. Given k with a < k < n? 2, suppose that all D j, where k < j n? 2, travel in the same direction, but D k travels in the other direction. Assume without loss of generality that D k travels north and all D j with j > k travel east. Then, letting = (k + 1; 0) 2 S k+1, we see that H contains the cycle [](E 2(n?k)?3 N). This contradicts the fact that H is a path.

8 8 MARGARET H. FORBUSH ET AL. The preceding proposition shows that the \middle" diagonals (that is, those D j with a < j < b) must all travel in the same direction. Theorem 3.5 shows that the other non-terminal diagonals have no eect on whether a spanning quasi-path is a hamiltonian path or not. First, we prove a lemma Lemma. Let H be a spanning quasi-path, with terminal diagonal S a [ S b (where a < b). Let (p; q) 2 S b+1, and let P be the unique path in H that starts at (p; q) and ends in S a, without passing through S a. Then the terminal vertex of P is (n? 1? p; n? 1? q), which is the inverse of (p; q). Proof. Let us begin by showing that P steps o the edge of the board exactly once. Since steps that do not go o the edge of the board move from S k to S k+1, and a < b + 1, it is clear that P steps o the edge of the board at least once. Let be the square from which P steps o the edge of the board (the rst time). If 2 S k, then the square immediately immediately after on P belongs to S 2n?2?k. Since k b + 1 = 2n? 2? a, we have 2n? 2? k a. Therefore, because a < n? 1, we know that P reaches S a before it reaches S n?1, which is the rst subdiagonal from which it is possible to step o the edge of the board. Let [ 0 ; 1 ; ; m ] be the sequence of squares visited by P. We now show that m is odd, that P steps o the edge of the board from (m?1)=2, and that, for 0 i m? 1, the squares i and m?1?i lie on the same diagonal. If P steps o the edge of the board from k, then k 2 S b+1+k, so k+1 2 S 2n?2?(b+1+k). On the other hand, because m 2 S a (and P does not step o the edge of the board between k+1 and m ), we must have k+1 2 S a?(m?k?1). Therefore, 2n? 2? (b k) = a? (m? k? 1): We know, from Proposition 2.4, that a + b = 2n? 3, so this implies m = 2k + 1, which means that m is odd, and that P steps o the edge of the board from (m?1)=2. Furthermore, for i < k, we have i 2 S b+1+i and m?1?i 2 S a?1?i. Since a + b = 2n? 3, we have (b i) + (a? 1? i) = 2n? 3, which means that i and m?1?i lie on the same diagonal, as desired. We are now ready to prove that the terminal square of P is (n? 1? p; n? 1? q). Assume, without loss of generality, that (m?1)=2 travels east. From the preceding paragraph, we know that P steps o the edge of the board from (m?1)=2, so (m?1)=2 must be of the form (m?1)=2 = (n? 1; y). Thus, exactly n? 1? p of the squares 0 ; 1 ; : : : ; (m?3)=2 travel east. Since i and m?1?i are on the same diagonal, then exactly n? 1? p of the squares (m+1)=2 ; (m+3)=2 ; : : : ; m?1 travel east. Similarly, exactly y? q of the squares (m+1)=2 ; : : : ; m?1 travel north. Because (m?1)=2 travels east, o the edge of the board, we have (m+1)=2 = (0; n? 1? y). Then n? 1? p steps east and y? q steps north result in the square (n? 1? p; n? 1? q), as desired Theorem. Let H be a hamiltonian path in B n, with terminal diagonal S a [S b (where a < b), and let H 0 be any spanning quasi-path in B n, with the same initial and terminal squares as H and such that each non-terminal diagonal D j, with a < j < b, travels exactly the same way in H 0 as it does in H. Then H 0 is also a hamiltonian path. Proof. From Proposition 2.10, we see that each square in the terminal diagonal travels exactly the same way in H 0 as it does in H. To show that H 0 is a hamiltonian path, it suces to show that H 0 has no cycles.

9 HAMILTONIAN PATHS IN PROJECTIVE CHECKERBOARDS 9 Suppose that there is a cycle C 0 in H 0. This cycle cannot be contained in H, so it is not dicult to see that it must contain at least one square in S b+1. We may write C 0 as the concatenation C 0 = P 1 + Q P r + Q 0 r of paths, such that each P i is contained in H \ H 0, and each Q 0 j starts in S b+1 and ends in S a, without passing through S a. Lemma 3.4 shows that, for each j, there is a path Q j in H that starts in S b+1 and ends in S a, without passing through S a, and has the same endpoints as Q 0 j. Then the concatenation C = P 1 + Q P r + Q r is a cycle in H. This is a contradiction Corollary. If there is a hamiltonian path H from an initial square to a terminal square, then there is a hamiltonian path H 0 from to, such that all non-terminal diagonals travel in the same direction. Proof. We may write the terminal diagonal of H in the form T = S a [ S b, with a < b. We know from Proposition 3.3 that all D j where a < j < b travel in the same direction. Assume without loss of generality that they travel east. Let H 0 be the spanning quasi-path of B n in which every non-terminal diagonal travels east, and each square in the terminal diagonal travels exactly the same way as it does in H. Theorem 3.5 shows that H 0 is a hamiltonian path from to. We now embark on the proof of Main Theorem Thus, our goal is to describe precisely which squares can be joined by a hamiltonian path (in which all non-terminal diagonals travel east). To streamline the required case-by-case analysis, we start with a few lemmas. These preliminary results provide helpful information on the locations of initial and terminal squares of hamiltonian paths Lemma. Let T be the terminal diagonal of a hamiltonian path from to, let be the square on T + with y = n? 1, and let be the square on T? with y = 0. (1) Suppose 2 T + and y < n? 1. If some square 2 T? travels east, then both and travel east. (2) Suppose N?1 2 T?. If some square in T + travels east, then both and travel east. (3) Suppose 2 T?. If some square in T + travels north, then travels north. (4) Suppose E?1 2 T +. If some square in T? travels north, then travels north. Proof. We prove only (1), because the other cases are similar. In the notation of Corollary 2.12, it is not dicult to see that u() = n? 1? y = u()? 1 < u() = u()? y u(): Since travels east, we know from Corollary 2.12 that u() < u(e?1 ), so this implies u(); u() < u(e?1 ), so both and travel east Lemma. Let H be a hamiltonian path in B n from to, such that every nonterminal diagonal travels east. Assume n is odd. Let T be the terminal diagonal of H, and let be the square on T + with y = n? 1. Then either: (1) y = 6n; or (2) y = 6n; or (3) travels north, and there is a square on T? with y = 6n, such that = EN?1 ; or

10 10 MARGARET H. FORBUSH ET AL. (4) travels north, and there is a square on T + with y = 6n, such that = N. Proof. Let be the square on T + with y = 6n, if such a square exists; otherwise, let = (n? 1; 6n). Let be the square on T? with y = 6n, if such a square exists; otherwise, let = (0; 6n). We may assume y 6= 6n and y 6= 6n; in particular, f; g \ f; ; E; Eg = ;. Case 1. travels east. Because 6=, the square must travel north, for, otherwise, H contains the cycle [](E n ). (Note that, because all non-terminal diagonals travel east, this implies 2 T +.) Since 6= E, we see from Proposition 2.10 that EN?1 must also travel north. Therefore 6= (n? 1; 6n), for, otherwise, H contains the cycle [](NENE n?1 ). Therefore n? x? 2 0, so (n? x? 2; 6n) is a square of B n. Then Proposition 2.4 implies (n? x? 2; 6n) 2 T?, so we have 2 T?. We must have EN?1 =, for, otherwise, Proposition 2.10 implies that EN?1 travels east, which would mean that H contains the cycle [](NE n+1 NE n?1 ). Hence Lemma 3.7(3) implies conclusion (3) holds. Case 2. travels north. (Note that 2 T?, so Proposition 2.4 implies (n? x? 2; 6n) 2 T +, which means 2 T +.) Since 6= E, we see from Proposition 2.10 that EN?1 travels north, as does. Then, since 6=, we see that must travel east, for, otherwise, H contains the cycle [E](E x?x?1 NE n?x+x+1 N): However, NE?1 does not travel east, for, otherwise, H contains the cycle [](NE n+1 NE n?1 ): Therefore, Proposition 2.10 implies N =. Hence Lemma 3.7(4) implies conclusion (4) holds Lemma. Let H be a hamiltonian path in B n, such that every non-terminal diagonal travels east. Assume n is even. Let T be the terminal diagonal of H, let be the square on T + with y = n? 1, and let be the square on T? with y = 0. Then either: (1) y = 6n and x 6n? 1; or (2) y = 6n? 1 and x 6n; or (3) y = 6n and x 6n? 1; or (4) y = 6n? 1 and x 6n; or (5) T 6= D n?1, and both and travel east. Proof. Let be the square on T + with y = 6n, if such a square exists; otherwise, let = (n? 1; 6n). Let be the square on T? with y = 6n? 1, if such a square exists; otherwise, let = (0; 6n? 1). We may assume f; g \ f; g = ;, for, otherwise, one of conclusions (1){(4) holds. Case 1. travels north. Note that we must have 2 T?, because all nonterminal diagonals travel east. Then we conclude from Proposition 2.4 that (n? 2? x ; 6n) 2 T +, so 2 T + and x < n? 1. The square must travel north, for, otherwise, H contains the cycle [](NE n ). We may assume that EN?1 travels north, as does, for, otherwise, we see from Proposition 2.10 that = E, which implies that conclusion (2) holds. Similarly,

11 HAMILTONIAN PATHS IN PROJECTIVE CHECKERBOARDS 11 we may assume that EN?1 travels north, as does, for, otherwise, we must have = E, which implies that conclusion (1) holds. Hence, H contains the cycle [](NE x?x NE n?x+x+1 NE x?x NE n?x+x?1 ): Case 2. travels east. Let 0 be the square on T + with 0 y = 6n? 1, if such a square exists; otherwise, let 0 = (n? 1; 6n? 1). Suppose 0 =. (Note that Proposition 2.4 then implies (n?1? 0 x ; 6n?1) 2 T?, so 2 T?.) We may assume T 6= D n?1, for, otherwise, this implies that = (6n; 6n? 1), so conclusion (4) holds. Furthermore, we may assume that 0 NE?1 travels east, for, otherwise, Proposition 2.10 implies = 0 N, in which case conclusion (1) holds. Therefore, Lemma 3.7(1) implies that and also travel east, so conclusion (5) holds. We may now assume 0 6=. Then 0 must travel east, for, otherwise, H contains the cycle [ 0 ](NE n ). Suppose NE?1 does not travel east. Then 2 T?, and, because travels east, we must have = N. Therefore, we may assume T 6= D n?1, for, otherwise, we have = (6n? 1; 6n), so conclusion (1) holds. Since 2 T?, Proposition 2.4 implies (n? 1? x ; 6n? 1) 2 T +, so 0 2 T +. Therefore, Lemma 3.7(2) implies that conclusion (5) holds. We may now assume NE?1 travels east. Furthermore, we may assume that 0 NE?1 travels east, as 0 does, for, otherwise, we see from Proposition 2.10 that 0 N =, which implies that conclusion (1) holds. Also, we may assume that 0 N travels north, for, otherwise, either conclusion (3) holds (if 0 N = ) or H contains the cycle [](E 2n ) (if 0 N travels east). Because all non-terminal diagonals travel east, this implies 0 N 2 T +, so 0 N =. Then it is clear that 0 62 T +, so we must have 0 = (n? 1; 6n? 1), which means = (n? 1; 6n). Therefore, E = 6=, so Proposition 2.10 implies that N?1 = EN?1 travels north, as does. Therefore, H contains the cycle [](E 2n?1 NEN) Lemma. Let H be a hamiltonian path in which all non-terminal diagonals travel east, and = (n? 1; 0). If n is odd, then 2 f(6n; 6n); (6n; 6n + 1)g. If n is even, then 2 f(6n? 1; 6n); (6n; 6n)g. Proof. We consider two cases. Case 1. n is odd. Let = (6n? 1; 6n) and = (6n; 6n). We have y = 0 6= 6n and 6= EN?1, so we see from Lemma 3.8 that either y = 6n or = N = (6n; 6n + 1). Thus, we may assume y = 6n, and we wish to show = (6n; 6n). Because 2 D n?1, we see from Corollary 2.11 that N?1 2 D n?1 = S n?2 [ S n?1. Thus, 2 f(6n; 6n); (6n + 1; 6n)g. If = (6n + 1; 6n), then H contains the cycle [E?1 ](NE n N). Case 2. n is even. If n = 2, the desired conclusion may be veried directly, so we may assume n > 2. Since y = 0 6= 6n, and x = n? 1 > 6n, and T = D n?1, we see that either (1) or (2) of Lemma 3.9 holds. Then, since we must have N?1 2 D n?1, we conclude that 2 f(6n; 6n); (6n? 1; 6n); (6n; 6n? 1)g. However, if = (6n; 6n? 1), then H contains the cycle [E?1 ](NE n ). By considering the inverses of these hamiltonian paths, we obtain the following corollary.

12 12 MARGARET H. FORBUSH ET AL Corollary. Let H be a hamiltonian path in which all non-terminal diagonals travel east, and = (0; n? 1). If n is odd, then 2 f(6n; 6n); (6n; 6n? 1)g. If n is even, then 2 f(6n? 1; 6n? 1); (6n; 6n? 1)g Main Theorem. Let and be two squares on the nn projective checkerboard B n, with n 3. There is a hamiltonian path from to in which all nonterminal diagonals travel east if and only if is in the diagonal containing N?1 and either: (1) n is odd and: (a) y = 0 and x 6n and y = 6n; or (b) = (0; n? 1) and 2 f(6n; 6n); (6n; 6n? 1)g; or (c) y = n? 1 and x 6n and y = 6n; or (d) = (n? 1; 0) and 2 f(6n; 6n); (6n; 6n + 1)g; or (2) n is even and: (a) y = 0 and x 6n? 1 and either: (i) y = 6n? 1 and x 6n? 1; or (ii) y = 6n and x 6n? 1; or (b) = (0; n? 1) and 2 f(6n? 1; 6n? 1); (6n; 6n? 1)g; or (c) y = n? 1 and x 6n and either: (i) y = 6n and x 6n; or (ii) y = 6n? 1 and x 6n; or (d) = (n? 1; 0) and 2 f(6n? 1; 6n); (6n; 6n)g. Furthermore, such a hamiltonian path is unique, if it exists. Proof. If and are as specied, then the desired hamiltonian path from to either appears explicitly in Figure 3 or Figure 4 (in cases 1a, 1b, 2a, and 2b), or is the inverse of a hamiltonian path in Figure 3 or Figure 4 (in cases 1c, 1d, 2c, and 2d). The uniqueness follows from Corollary Now suppose there is a hamiltonian path H from to, such that all nonterminal diagonals travel east. Let T be the terminal diagonal of H, let be the square on T + with y = n?1, and let = (n?1; 0). From Corollary 2.11, we know that N?1 2 T. From Lemmas 3.8 and 3.9, we see that it is not possible to have both = and y = 0, so we may assume 6=. (If =, replace H with its inverse ~ H. Note that if ~ H satises condition 1a or 2a, then H satises condition 1c or 2c, respectively.) Case 1. T 6= D n?1. The square must travel east, for, otherwise, H contains the cycle [](NE 2x+1 ). Let be the square on T? with y = 0. We see that does not travel east, for, otherwise, H contains the cycle [](E 2n ). Then, since EN?1 = travels east, but does not, we conclude from Proposition 2.10 that = E. (In particular, y = 0.) Suppose n is odd. Then, because does not travel north, we see from Lemma 3.8 that y = 6n. Therefore, since 0 x n?1, we have x + y 6n and 2n?3?( x + y ) 6n? 1. Then, since 2 T, and y = 0, Proposition 2.4 implies x 6n? 1, so x 6n. Hence, conclusion 1a holds. Now suppose n is even. Note that 62 f(6n; 6n? 1); (6n? 1; 6n); (6n? 1; 6n? 1)g;

13 HAMILTONIAN PATHS IN PROJECTIVE CHECKERBOARDS 13 because T 6= D n?1. Then, because y = 0 =2 f6n; 6n? 1g and does not travel east, we see from Lemma 3.9 that either (1) y = 6n and x 6n; or (2) y = 6n? 1 and x 6n? 2. Therefore, x + y 6n? 1 and 2n? 3? ( x + y ) 6n? 2. Then, since E?1 2 T (and y = 0), Proposition 2.4 implies x?1 6n?2. Therefore, conclusion 2a holds. Case 2. T = D n?1 and travels north. Note that = (0; n? 1). We know that does not travel east, for otherwise H contains the cycle [](NE). If =, then Lemma 3.10 implies that conclusion 1d or 2d holds. We may now assume travels north. We see immediately that N?1 does not travel north, for, otherwise, H contains the cycle [](NNEN). Thus, E?1 = travels north, but N?1 does not, so Proposition 2.10 implies =. Hence, Corollary 3.11 implies that conclusion 1b or 2b holds. Case 3. T = D n?1 and travels east. We may assume 6=, for, otherwise, Lemma 3.10 implies that conclusion 1d or 2d holds. Therefore, travels either north or east. Suppose =. If n is odd, then Lemma 3.8 implies that conclusion 1a holds, whereas, if n is even, then Lemma 3.9 implies either that conclusion 2a holds, or that = (6n; 6n? 1). (If n is even, note that 6= (6n; 6n), because (6n; 6n) =2 D n?1.) However, in the latter case, H would contain the cycle [E?1 ](NE n ). Thus, we henceforth assume 6=. Suppose travels east. Let = E?1. Since the square EN?1 = travels east, and E = 6=, we see from Proposition 2.10 that that also travels east. Thus, H contains the cycle [](E 2n ). We may now assume travels north. Furthermore, we may assume 6=, for, otherwise, Corollary 3.11 implies that conclusion 1b or 2b holds. Then, since travels north, and E = 6=, Proposition 2.10 implies that EN?1 also travels north. Similarly, since travels east, and N = 6=, it must be the case that NE?1 also travels east. Thus, H contains the cycle [](E 2n?1 NEN). Acknowledgments. This research was partially supported by grants from the National Science Foundation and the New England Consortium for Undergraduate Science Education. Many helpful comments on the manuscript were received from Stephen Curran and an anonymous referee. References 1. S. J. Curran and D. Witte, Hamilton paths in cartesian products of directed cycles, Ann. Discrete Math. 27 (1985), 35{ J. A. Gallian and D. Witte, When the cartesian product of directed cycles is hyperhamiltonian, J. Graph Theory 11 (1987), 21{ D. Housman, Enumeration of hamiltonian paths in Cayley diagrams, Aequationes Math. 23 (1981), 80{ D. S. Jungreis, Generalized hamiltonian circuits in the cartesian product of two directed cycles, J. Graph Theory 12 (1988), 113{ L. E. Penn and D. Witte, When the cartesian product of directed cycles is hypohamiltonian, J. Graph Theory 7 (1983), 441{ W. T. Trotter and P. Erdos, When the cartesian product of directed cycles is hamiltonian, J. Graph Theory 2 (1978), 137{142.

14 14 MARGARET H. FORBUSH ET AL. 7. A. M. Wilkinson, Circuits in Cayley digraphs of nite abelian groups,, J. Graph Theory 14 (1990), 111{116. Department of Mathematics, Williams College, Williamstown, MA Department of Mathematics, Virginia Military Institute, Lexington, VA Department of Mathematics, Williams College, Williamstown, MA Department of Mathematics, Williams College, Williamstown, MA Department of Mathematics, Williams College, Williamstown, MA Department of Mathematics, Williams College, Williamstown, MA Department of Mathematics, Middlebury College, Middlebury, VT Department of Mathematics, Williams College, Williamstown, MA Department of Mathematics, Williams College, Williamstown, MA Current address: Department of Mathematics, Oklahoma State University, Stillwater, OK address: