Snappea Tutorial This tutorial walks you through the most useful features of the 'snappea' (pronounced "snap pea", like the vegetable) hyperbolic structures programs. [ UNIX users: snappea sometimes writes things out to files, and it always does this relative to the directory from which you invoke it. Sometimes the files go into subdirectories of the current directory (with names like "generators", "Dirichlet_domains", ...), and snappea will create these subdirectories if they don't yet exist. If you plan to use snappea very much it might be a good idea to create a snappea "work directory" to work in, and then always invoke snappea from that directory. ] (0) If you are using a Mac, start the program by double-clicking on the program icon. If you are using a UNIX workstation, start it by typing "snappea" at the shell prompt. You should get a "select program" prompt. Snappea is really an umbrella program which gives you access to several constituent programs. (Eventually there'll be a smoother interface for all this.) (1a) If you are using a Mac, type "l" (that's the letter "el") to start the link entry program. If you are not using a Mac, skip down to step 2; the link entry program is not yet available on other systems. (1b) Use the mouse to draw the figure eight knot. Click to make the vertices. To close a component, click on the initial vertex. By the way, if you attempt to add a segment that's not in general position relative to what you've already drawn the program will beep at you and refuse to add that segment. You can now adjust the crossings (over vs. under) by clicking on them, or you can choose Alternate from the Edit menu to fix them all at once. (1c) Choose Save... from the File menu to save the link you have drawn. Give it a one-word name like "fig8". First the program saves the projection in a folder called "link projections" (which it creates if necessary), then it triangulates the complement, finds a hyperbolic structure for it, and saves the result in the folder "manifolds" (which it also creates if necessary). It tells 1you something about the hyperbolic structure; when you're done looking at the information, click the mouse to return to the main link entry program. If you want, you can play around with the other menu commands, which are all self-explanatory. [Note: when you save link complements the manifold's cusps will be numbered in the same order you drew the corresponding link components.] (1d) Choose Quit Link Entry from the File menu. (2a) Type "d" to start the Dehn filling program. At the "which manifold" prompt, type "fig8" (or whatever you called your figure eight knot complement). If you were not able to do step 1 because you are not using a Mac, type "figure8" which will cause snappea to load the figure eight knot complement from its supplied library of manifolds. You'll see some information about this manifold. (2b) If you are on a Mac, or an IRIS with geomview, type "dd" (for Dirichlet domain) and after a brief computation you'll see a Dirichlet domain for the figure eight knot complement. [ On an IRIS it may take several seconds for geomview to appear. ] Drag it with the mouse to rotate it. If you let go of the mouse button while the mouse is still moving the Dirichlet polyhedron will keep spinning. On Macs: hit Command-S for a see-through Dirichlet domain. Click the close box of the viewing window when you're done playing with the Dirichlet domain. On IRISes with geomview: you can use all the features of geomview to view and manipulate the Dirichlet domain. See the geomview documentation for details. On UNIX workstations without geomview: typing "dd" will cause snappea to write an OOGL file containing the Dirichlet domain. Snappea will prompt you for the name of this file. You can then view this file with geomview on an IRIS somewhere if you have access to one. On any computer: if you type "ddf" instead of "dd", snappea will prompt you for the name of a file and will write an OOGL data file by that name containing the Dirichlet domain. (2c) Type "5 1" to ask snappea to do (5, 1) Dehn filling on the figure eight knot complement. You'll see the volume and homology of this closed hyperbolic 3-manifold, and also the length and torsion of the core geodesic of the filled in solid torus. Type "dd" to see the Dirichlet domain (the computation will be a little slower this time (particularly on a MacPlus or SE)). Note the tetrahedral symmetry. The program finds Dirichlet domains centered at points of locally maximal injectivity radius. (2d) Type "com" to return to the complete hyperbolic structure for the figure eight knot complement. Type "cs" to toggle the Chern- Simons calculator on. You'll be asked for the Chern-Simons invariant of the present configuration. The present configuration is the complete structure on the figure eight knot complement, and its Chern-Simons invariant is 0, so type 0. Then type "5 1" again, and this time the Chern-Simons invariant will be displayed along with the other information. (2e) Type "gen" to create a set of generators (in PSL(2,C)) for the fundamental group of this manifold. (This will take several seconds.) A file containing the generators will be created automatically and saved in a folder/directory called generators under the name "fig8(5,1)". (2f) Type "-5 1" and you'll see that (-5,1) Dehn filling has the same invariants as the (+5,1) Dehn filling. Are they isometric? Type "gen" to create the generators for the (-5,1) filling, and then type "diff" to compare the two files (you just hit return twice to have diff use the two most recently computed sets of generators, which is what we want). You'll see that the lists of generators are complex conjugate, which means that the (+5,1) filling and (-5,1) filling are isometric, but oppositely oriented. Technical comments (optional): When the generators are different you can't conclude that the manifolds are nonisometric, because the generators might have been computed relative to different basepoints. The basepoint is always a local maximum of the injectivity radius function, but there may be several such local maxima. In pratice this is seldom a problem. (Also, one can obtain different basepoints by using different starting points in the basepoint finding algorithm, but I still haven't settled on a good user interface for this.) When two manifolds are not isometric you have to use other invariants to distinguish them. For manifolds obtained by doing several Dehn fillings on a multicusp manifold the Dirichlet domain program occasionally fails due to the excessive roundoff error which accumulates during the computation of the canonical basepoint. There's a compile-time option to use larger error tolerances (the drawback to larger error tolerances is that when you go to do high-order Dehn fillings, which yield Dirichlet domains with lots of tiny faces, some of the tiniest faces get swallowed up entirely by the error tolerances). Get in touch with me if you have problems with these algorithms not working for the cases you are interested in. Note: the Dehn fillings needn't be on the same cusped manifold. For example, the program will show you that m003(-1,3) and m004(5,1) are isometric. ("m003" and "m004" will be explained below) (2g) Type "complete" (or just "com") to return to the complete (i.e. not Dehn filled) hyperbolic structure on the figure eight knot complement. At the ">" prompt type "ls" to compute a complex length spectrum ("complex" means it includes both the length and the torsion for each closed geodesic included). Type "2.5" for "tiling radius" and 2.0 for "cutoff length". After a short computation (or a long computation if you're using a MacPlus) you'll be prompted for a file name under which to store the list of lengths and torsions of closed geodesics up to length 2.0. The file will be placed in a directory called "length spectra", which the program creates if necessary. You may examine this file using any word processor (e.g. the one you use to read this tutorial). Note that because the figure eight knot complement has an orientation reversing symmetry each closed geodesic has a mate of the same length but opposite torsion. If you compute a length spectrum for the (5,1) Dehn filling (by typing "5 1", then "ls", "2.5", "2.0" and a file name) you'll see that the lengths do not occur in matched pairs, which implies that the manifold is chiral. New feature: typing "tube" at the ">" prompt will compute a tube radius of each geodesic. I recommend a radius of about 2.0. Warning: the tube radius algorithm is not carefully thought out, so I don't guarantee its correctness. Technical comments on choosing values for the "tiling radius" and the "cutoff length". You should probably just skip this material and go on to paragraph (2h) below, but if you are particularly interested in length spectra, then read on. The program computes length spectra by tiling hyperbolic space with lifts of a (hypothetical) fundamental domain. To decide when to stop tiling, it keeps track of how far each group element translates the basepoint. If the basepoint is translated a distance less than the "tiling radius", the program continues its recursive tiling by considering the lifts obtained by multiplying the given group element by each generator in a precomputed generating set. When the recursion is complete the program computes the complex length associated to each group element it found, and reports the lengths whose real part (i.e. the actual length, ignoring torsion) is less than the "cutoff length". Obviously the "cutoff length" should be less than the "tiling radius". In practice, "tiling radius" = "cutoff length" + 0.5 works well. If you want to verify the result you can run it again with "tiling radius" = "cutoff length" + 1.0. Using "cutoff length" = 3.0 and tiling radius = 3.5 gives fast results; "cutoff length" = 3.5 and "tiling radius" = 4.0 is also tolerably fast. The number of group elements increases exponentialy with the distance, of course, so you can go a little beyond that (say to 5.0) but not a whole lot further. If you want to check out the accuracy more systematically (e.g. if you find a particularly interesting set of spectra) I recommend the following procedure. Do, say, "tiling radius" = "cutoff length" = 3.5, then do "tiling radius" = "cutoff length" = 4.0 and see where the first new length occurs. You can be fairly certain that the spectrum is complete up to that point. Finding a delta such that the list of lengths up to the "cutoff length" is guaranteed to be complete when the "tiling radius" is greater than "cutoff length" + delta is tricky, even for a fixed "cutoff length". The catch is that is is theoretically possible that the recursion could terminate (having reached lifts all of which translate the basepoint a distance greater than "tiling radius") and still miss some lifts which are further away in a combinatorial sense yet nevertheless translate the basepoint a distance less than the "tiling radius". (In practice this will occur seldom if at all.) Fortunately it is still possible to get some bounds which will rigorously guarantee the completeless of the reported length spectrum. In the case of a closed manifold with fundamental domain of radius delta, one can prove that the reported length spectrum will be complete when "tiling radius" > "cutoff length" + 2*delta. Unfortunately this bound isn't much good for manifolds with short geodesics (because delta will be large, which makes the prescribed "tiling radius" intractibly large) and for cusped manifolds it's useless. Other techniques are possible in the latter cases (namely getting bounds on how far a closed geodesic of length less than "cutoff length" can travel up the cusp) but they are messy and I haven't pursued them. (2h) Hit return at the ">" prompt to stop doing Dehn fillings on the figure eight knot complement, then hit return again at the "which manifold" prompt to quit the Dehn filling program. (3) There are five ways to create new manifolds. Each will be described below. For reference, here's the list of all five ways. (a) Use the link entry program [on the Mac only]. (b) Recall manifolds from the "census" (i.e. data base) of all cusped hyperbolic 3-manifolds obtainable from 5 or fewer ideal tetrahedra. (c) Recall manifolds from the census of all orientable cusped hyperbolic 3-manifolds obtainable from 6 ideal tetrahedra. (d) Recall manifolds from the link library which contains all the knots and links which appear in the standard tables (e.g. Rolfsen). (e) Create punctured torus bundles. You can, of course, do Dehn fillings on manifolds obtained in any of the above ways. (3a) The link entry program was described above. (3b) Return to the Dehn filling program by typing "Dehn" (or just "d") and at the "which manifold?" prompt type "m003". This calls up a manifold from the so-called "census" of all 415 cusped hyperbolic 3-manifolds obtainable from 5 or fewer ideal tetrahedra. The census manifolds are named m000 through m414, and are listed in order of increasing volume (e.g. m000 is the Gieseking). A table giving various invariants for these manifolds is available. Note that even though the manifold m003 has the same volume as the figure eight knot complement, it has first homology Z + Z5, and is therefore not a knot complement. These "census manifolds" are my favorite source of examples because they are completely general and thus provide a better idea of what hyperbolic manifolds are like than one gets by, say, looking only at knot and link complements. Both orientable and nonorientable manifolds are included in the census. The census manifolds are stored in a "dehydrated" form and are "rehydrated" on the fly whenever you need to use one. That's why you get the message about "rehydrating m003... solution found...". (3c) A similar library exists for the 962 orientable manifolds obtainable from 6 (but not fewer) ideal tetrahedra. They are named "s000" through "s961". Similarly, the 3552 orientable manifolds obtained from 7 (but not fewer) ideal tetrahedra are named "v000" through "v3551". (3d) Hit return at the ">" prompt to close the m003 manifold, and at the "which manifold" prompt type "L306002". This gives you the Borromean rings complement. All the knots and links from the standard tables are available, as well as 11-crossing knots and 10-crossing links (thanks to Joe Christy). The n-th knot or link of m crossings and p components is named "Lpmmnnn" (e.g. the Borromean rings are the second 6-crossing 3-component link). You can do Dehn fillings on the Borromean rings like you did on the figure eight knot complement. The program will start you off doing Dehn fillings on cusp number 0. To change to one of the other cusps, type "cusp 1" or "cusp 2". At any point you may view the shapes of the tetrahedra (given as complex edge parameters) and the Dehn filling coefficients of the cusps by typing "view". For a complete listing of the commands available in the Dehn_filling program, type "help" at the ">" prompt. When you're done with the Borromean rings hit return at the ">" prompt and again at the "which manifold" prompt to leave the Dehn filling program. Note: if you want to rehydrate all the links in the link library and save them in the manifolds folder/directory (so you don't have to rehydrate them on the fly) type "r" at the "select program" prompt. This operation takes a while, so you probably don't want to do it now. (3e) At the "select program" prompt type "bundles" (or just "b") to enter the punctured torus bundle program. Type "+" for the sign of the determinant, "+" for the sign of the trace, and "RL" for the RL factorization, and you will see that you've created a manifold of the same volume (2.02...) as the figure eight knot complement and the manifold m003. The information you provided gives a factorization of the monodromy of the desired bundle. The string of R's and L's provides a factorization in terms of the 2 x 2 matrices R = (1,1;0,1) and L = (1,0;1,1). In addition, if you specify "-" for the determinant the result will be multiplied by (0,1;1,0), and if you specify "-" for the trace the result will be multiplied by (-1,0;0,-1). (Note: the string of R's and L's read left to right determines the sequence of matrices read right to left.) Type "+" for the determinant, "-" for the trace, and RL for the RL-factorization. You'll get another manifold of volume 2.02... . Hit return to leave the bundle program. Warning: At the moment the meridian and longitude on the cusp of a punctured torus bundle are the ones natural relative to the Euclidean structure on the cusp (namely the shortest and second shortest curves), NOT the usual meridian and longitude one expects to find on a punctured torus bundle. In the near future I will write the code to put on the ones natural to the bundle. (4) Go back to the Dehn filling program and examine the manifolds "B++RL" and "B+-RL" you just created, and also the census manifolds m003 and m004. An examination of the various invariants for these manifolds suggests that B++RL is isometric to m004 and B+-RL is isometric to m003. To prove they are isometric you can ask snappea to compute a canonical triangulation for each and then check whether the canonical triangulations are combinatorially the same. To do this, hit return to leave the Dehn filling program, and then type "isometric" (or just "i") to use the isometry checker. Provide the names B++RL and m004 and snappea will tell you they are isometric; Provide B++RL and m003 snappea will tell you they're different. While you're at it you can show that B++RL = m004 = L104001 = fig8. When you're done hit return to leave the isometry checker. The canonical triangulations are dual to the Ford domain. In rare cases the cell-division dual to the Ford domain contains cells other than tetrahedra, and in these cases there is no unique canonical triangulation. Snappea will report this ambiguity and refuse to speculate on whether the manifolds are the same (unless they happen to have the same triangulations to begin with). I hope to correct this deficiency in the future. Note: There are two ways to check whether two manifolds are isometric: you can check canonical triangulations as just described, or you can compute generators for the group of covering transformations as described in (2f) above. Comparing canonical triangulations has several advantages: (i) It's faster. (ii) It works in virtually all cases. (iii) In the final analysis it is comparing two combinatorial triangulations, so there is absolutely no possibility of round-off error leading to a "false positive" result. By contrast, two very high order Dehn fillings on a cusped manifold will have nearly identical generators. (In fact, comparision of generators MUST fail to resolve some such pairs of manifolds because any finite computer can represent only a finite number of distinct matrices, yet there are infinitely many distinct high order Dehn fillings.) It has one disadvantage, namely that it can be used only for the original cusped hyperbolic 3-manifolds, not for the manifolds represented as Dehn fillings. (5) If you are not using a Mac, skip this step. If you are using a Mac, at the "select program" prompt type "p" to use "pretty pictures", and at the "which manifold" prompt type "L306002" to get the Borromean rings. A couple messages with advice for printing hard copies will appear; you can safely ignore them for now. Type "0" to take a look at cusp number 0. You'll be asked for a lower bound on the diameter of the horoballs you want plotted. On a Mac II or SE/30 (or any newer Mac with a floating point coprocessor) 0.2 is a reasonable choice. If you've got a MacPlus or SE, try 0.3 or 0.35. (The number of horoballs grows outrageously fast as the minimum diameter decreases, so the difference in time and memory requirements between 0.3 and 0.2 is more than you might expect.) A blank screen with a menu bar appears. (If you're lucky enough to have a color monitor you should, at this point, choose the Color Monitor option on the Options menu. If you have THINK C you can later change a #defined constant to make the Color Monitor option on by default.) (In the new SnapPea the program will take care of these matters automatically.) Choose the Horoballs item from the Draw menu. A picture will appear which shows the view in the manifold as seen looking down cusp 0 "from infinity". The view in the manifold is, of course, the same as the view in the universal cover. Each disk you see is a lift to the universal cover of a horospherical neighborhood of one of the cusps. The program chooses the largest horospherical neighborhoods which have equal volumes and disjoint interiors. Different fill patterns (or different colors on a color monitor) distinguigh the different cusps. (You can choose Fill Patterns from the Draw menu to see which pattern/color corresponds to which cusp.) You are, of course, looking at the manifold from inside one of the horospherical neighborhoods; we imagine its horospherical boundary is opaque from the outside but transparent from the inside so that it doesn't hide the other horoballs. A more prosaic interpretation of the horoball picture is that it's the universal cover of the manifold as represented in the upper half space model of hyperbolic 3-space, viewed from above, with the boundary of the current cusp represented as a horizontal plane at height one (the horizontal plane at height one is not explicitly shown in the picture to avoid hiding the other horoballs). Play around with the various Draw menu commands and Options menu options, and when you are done choose Quit Drawing to return to the console. You may now draw the pictures for cusps 1 and 2 if you wish (but because the Borromean rings is so symmetrical the pictures will be the same as for cusp 0, only with the fill patterns permuted). When you're done with the Borromean rings hit return in response to "which cusp". If you want to look at other examples, the first ten or twenty census manifolds (m000 - m020) offer some nice patterns. It's also interesting to create a punctured torus bundle whose RL-factorization is almost but not quite periodic, like B+-RLRLLLRLRLRLRLRLLRLRLRL. (Note: the console window supports copy-and-paste, so you never have to retype those long bundle names by hand. I.e. copy the name from the bundle program and paste it into the pretty pictures program at the "which manifold" prompt.) When you're done, hit return at the "which manifold" prompt to leave pretty pictures. (6) SnapPea can compute symmetry groups of manifolds. Type "symmetries" (or just "s") at the "select program" prompt, and enter the name of whatever manifold interests you. For example, you'll find that the symmetry group of m003 is Z/2 x Z/4, and the symmetry group of m004 is the dihedral group D4 (a group of order 8). SnapPea reports both the symmetry group of the manifold and the symmetry group of the corresponding link. For a manifold which you created as a knot or link complement to begin with, the symmetry group of the link is just what you expect it to be; in general, the symmetry group of the "link" is defined to be the subgroup of the full symmetry group consisting of those symmetries which take meridians to meridians in whatever coordinate system is employed on the boundary components of the manifold. (7) Last but not least, snappea can compute presentations for fundamental groups. At the "select program" prompt type "f" (for "fundamental group"), and when asked "which manifold" type "L306002" for the Borromean rings. You'll see a presentation for the fundamental group of the Borromean rings, followed by the words representing the peripheral curves. You are now in an environment similar to the Dehn filling environment. In response to the "pi one>" prompt, type "1 0" to do a (1,0) Dehn filling on cusp 0 of the Borromean rings. You will be told that the fundamental group of the Dehn filled manifold is a free group on two generators. Type "info" (or just "i") to see the status of all cusps. (By convention (0,0) means that a cusp is unfilled.) Try some other Dehn fillings on this cusp, and then return to the (1,0) Dehn filling before proceeding. Type "cusp 1" (or just "c 1") to switch to cusp 1. Type "1 0" to do a (1,0) Dehn filling on this cusp and you'll see that the fundamental group of the result is the integers. Type "info" to see the status of the cusps if you want. Now type "c 2" and "1 0" to switch to cusp 2 and do a (1,0) Dehn filling. Snappea will tell you that you have a homotopy sphere. Hit return when you're done. For more information on the programs available within snappea, type "help" at the "select program" prompt. By the way, the snappea console window on the Mac really is resizable, even though it has no resizing box and the cursor is always an I-beam! Just move the cursor to the lower right hand corner of the window, pretend there's a resizing box, and everything will work fine. Note: Many of the longer computations can be aborted by holding down control-C until the word "CANCEL" appears. I am eager to receive any and all comments and suggestions for improving these programs, from major additions to minor annoyances and typos. Please send comments and questions to weeks@poincare.geom.umn.edu or Jeff Weeks, Geometry Supercomputer Project, 1200 Washington Ave. S., Minneapolis, MN 55415. Thanks. CREDITS: Craig Hodgson has done large amounts of the programming, particularly in the routines to compute the Chern-Simons invariant and the quasicanonical generators for the groups, and in the enumeration of closed manifolds. More generally he has contributed to the design of the entire project from the very beginning. More recently, Joe Christy has made substantial contributions, and is coauthoring the GUI for the new SnapPea. The winged edge data structure in the Dirichlet domain program is from a Dirichlet domain program written by Charlie Gunn, which was the inspiration for the present one. Diane Hoffoss wrote the routine which generates lists of "nearby" group elements and computes length spectra. Martin Hildebrand did most of the work on the Census. Colin Adams, Bob Meyerhoff and Walter Neumann have also made significant contributions. Bill Thurston is responsible for most of the mathematics underlying these programs.
Comments to:
webmaster@www.geom.uiuc.edu
Created: Feb 08 1996 ---
Last modified: Jun 18 1996