1\newpage 2\section{Doing tropical computations} 3\label{sec:tropical} 4\emph{This section follows the max convention for tropical 5arithmetic. For the non-constant coefficient case tropical varieties are defined as in \cite{lifting} and \cite{thesis}.} 6\vspace{0.5cm} 7 8In this section we 9explain how to use Gfan to do tropical computations. For a fixed ideal $I\subseteq 10k[x_1,\dots,x_n]$ the set of all faces of all full-dimensional 11Gr\"obner cones is a polyhedral complex which we call the Gr\"obner 12fan of $I$. For tropical computations the lower dimensional cones of 13the complex will be of our interest. In general every Gr\"obner cone 14is of the form: 15$$C_\omega(I):=\overline{\{\omega'\in\R^n:\init_{\omega'}(I)=\init_\omega(I)\} 16}.$$ 17 18We define the tropical variety $\T(I)$ of an ideal $I$ to be the 19the set of all $\omega$ such that $\init_\omega(I)$ does not contain a monomial. 20If the ideal $I$ is homogeneous with respect to a positive grading, then the Gr\"obner cones cover all of $\R^n$ and $\T(I)$ is a union of Gr\"obner cones. 21Thus for a homogeneous ideal the tropical variety gets the structure of a polyhedral fan which it inherits from the Gr\"obner fan. 22%collection of all Gr\"obner cones with monomial-free initial 23%ideals. 24We therefore also define the tropical variety $\T(I)$ to be the collection of all Gr\"obner cones $C_\omega(I)$ such that $\init_\omega(I)$ is monomial-free. 25 26We start by noticing that for computational purposes it is no restriction to only consider the case of a homogeneous ideal: 27\begin{lemma}\cite[Lemma~6.2.5]{thesis} 28\label{lem:tropical by homogenisation} 29Let $I\subseteq k[x_1,\dots,x_n]$ be an ideal generated by 30$f_1,\dots,f_m\in k[x_1,\dots,x_n]$. Let $J=\langle 31f_1^h,\dots,f_m^h\rangle\subseteq k[x_0,\dots,x_n]$. Then $\init_\omega(I)$ is monomial-free if and 32only if $\init_{(0,\omega)}(J)$ is monomial-free where $\omega\in\R^n$. In particular we have the following identity of sets in $\R^{n+1}$: 33$$\{0\}\times\T(I)=\T(J)\cap(\{0\}\times\R^n).$$ 34\end{lemma} 35Here $f^h$ denotes the homogenization of the polynomial $f$. 36The homogenization of a list of polynomials can be computed by the program \texttt{gfan\_homogenize}. 37Notice that the lemma only requires the generators to be homogenized as a set of polynomials and not in the sense of a polynomial ideal. 38 39The tropical algorithms 40implemented in Gfan are explained in \cite{ctv}. 41% The reference also 42%contains definitions and theorems needed for understanding this 43%section of the manual. 44Notice that Gfan follows the usual conventions 45for signs of weight vectors defining initial forms while \cite{ctv} 46uses opposite signs. This means that \name is compatible with the max-plus convention whereas \cite{ctv} is compatible with the min-plus convention. 47 48\subsection{Tropical variety by brute force} 49The command \texttt{gfan\_tropicalbruteforce} will compute all 50Gr\"obner cones of a homogeneous ideal and for each check if its initial ideal 51contains a monomial. The output is the tropical variety of the 52ideal. Since the tropical variety is usually much smaller than the 53Gr\"obner fan this is a rather slow method for computing the tropical 54variety. The line 55\begin{verbatim} 56gfan_buchberger | gfan_tropicalbruteforce 57\end{verbatim} 58run on the input 59\begin{verbatim} 60Q[a,b,c,d,e,f,g,h,i,j] 61{ 62bf-ah-ce, 63bg-ai-de, 64cg-aj-df, 65ci-bj-dh, 66fi-ej-gh 67} 68\end{verbatim} 69produces a tropical variety of the input ideal in a few minutes as a 70polyhedral fan, see Section~\ref{format:fan}. We use 71\texttt{gfan\_buchberger} since \texttt{gfan\_tropicalbruteforce} 72requires its input to be a marked reduced Gr\"obner basis. 73\begin{remark} 74Notice that if $k'\supseteq k$ is a field extension and $I\subseteq 75k[x_1,\dots,x_n]$ an ideal then $\T(I)=\T(\langle 76I\rangle_{k'[x_1,\dots,x_n]})$ as a polyhedral fan. This identity 77follows since both objects can be computed by Gr\"obner basis methods 78and Gr\"obner bases are independent of such field extensions. The same argument of course also applies to the Gr\"obner fans of the two ideals. 79\end{remark} 80\subsection{Traversing tropical varieties of prime ideals} 81Let $I\subseteq \CC[x_1,\dots,x_n]$ be a homogeneous monomial-free 82prime ideal of dimension $d$. By the Bieri Groves Theorem \cite{bg} 83the tropical variety of $I$ is a pure $d$-dimensional polyhedral 84fan. It is connected in codimension one (\cite[Theorem~14]{ctv}) and can be 85traversed by Gfan. Let $\omega$ be a relative interior point of a 86$d$-dimensional Gr\"obner cone in the tropical variety of $I$. Fix 87some term order $\prec$. Gfan represents $C_\omega(I)$ by the pair of 88marked reduced Gr\"obner bases 89$(\G_{\prec_\omega}(\init_\omega(I)),\G_{\prec_\omega}(I))$. To 90compute the tropical variety of an ideal we must begin by finding a 91starting $d$-dimensional Gr\"obner cone. For this 92\texttt{gfan\_tropicalstartingcone} is used. % This programs guesses a 93%starting cone by heuristic methods. The guessing might fail. In that 94%case the program will terminate with an error message. 95After having 96computed a starting cone we use the program 97\texttt{gfan\_tropicaltraverse} to traverse the tropical variety. 98%There are several options for the output. In general the choice of output we request can have a huge influence on the running time for the program. 99We illustrate the procedure with an example. 100\begin{remark} 101Gfan does its computations over $\Q$ and thus the input should be an 102ideal generated by polynomials in $\Q[x_1,\dots,x_n]$. The assumption 103that $I$ is an ideal in $\CC[x_1,\dots,x_n]$ is needed since by 104``prime ideal'' in the above we mean ``prime ideal in the polynomial 105ring over the algebraically closed field $\CC$''. If $I$ is a prime ideal in 106$\Q[x_1,\dots,x_n]$ we do not know that its tropical variety is connected. 107 In 108Section~\ref{sec:non-constant} we address the problem of specifying non-rational 109coefficients. 110\end{remark} 111\begin{example} 112Let $I\subseteq\Q[a,\dots,o]$ be the ideal generated by the relations 113on the 2 by 2 minors of a 2 by 6 generic matrix. In $\CC[x_1,\dots,x_n]$ the 114ideal $I$ generates a prime ideal. To get a starting cone for the 115traversal of $\T(I)$ we run the command 116\begin{verbatim} 117gfan_tropicalstartingcone 118\end{verbatim} 119on the input 120\begin{verbatim} 121Q[a,b,c,d,e,f,g,h,i,j,k,l,m,n,o] 122{ 123bg-aj-cf, bh-ak-df, bi-al-ef, ck-bm-dj, ch-am-dg, 124cl-ej-bn, ci-eg-an, dn-co-em, dl-bo-ek, di-ao-eh, 125gk-fm-jh, gl-fn-ij, hl-fo-ik, kn-jo-lm, hn-im-go 126} 127\end{verbatim} 128and get a pair of marked reduced Gr\"obner bases 129\begin{verbatim} 130Q[a,b,c,d,e,f,g,h,i,j,k,l,m,n,o] 131{ 132l*m+j*o, i*m+g*o, i*k-h*l, i*j-g*l, h*j-g*k, 133e*m+c*o, e*k+b*o, e*j-c*l, e*h+a*o, e*g-c*i, 134c*k-b*m, c*h-a*m, b*i-a*l, b*h-a*k, b*g-a*j} 135{ 136l*m-k*n+j*o, i*m-h*n+g*o, i*k-h*l+f*o, i*j-g*l+f*n, h*j-g*k+f*m, 137e*m-d*n+c*o, e*k-d*l+b*o, e*j-c*l+b*n, e*h-d*i+a*o, e*g-c*i+a*n, 138c*k-d*j-b*m, c*h-d*g-a*m, b*i-e*f-a*l, b*h-d*f-a*k, b*g-c*f-a*j} 139\end{verbatim} 140This takes about a second. We store the output in the file \texttt{grassmann2\_6.cone} for later use. Since $I$ has many symmetries we add the following lines describing the symmetry group to the end of the file: 141\begin{verbatim} 142{ 143(0,8,7,6,5,4,3,2,1,14,13,11,12,10,9), 144(5,6,7,8,0,9,10,11,1,12,13,2,14,3,4) 145} 146\end{verbatim} 147We are ready to traverse $\T(I)$. 148% We start by running the program with the simplest possible output and using symmetries: 149We run the following command 150\begin{verbatim} 151gfan_tropicaltraverse --symmetry <grassmann2_6.cone 152\end{verbatim} 153The computation takes a few (two - three) minutes. The output looks like this: 154\begin{verbatim} 155_application PolyhedralFan 156_version 2.2 157_type PolyhedralFan 158 159AMBIENT_DIM 16015 161 162DIM 1639 164 165LINEALITY_DIM 1666 167 168RAYS 1690 0 -1 0 0 0 0 0 0 0 0 0 0 0 0 # 0 1700 0 0 0 0 0 0 -1 0 0 0 0 0 0 0 # 1 1710 0 0 0 0 0 0 0 0 0 0 -1 0 0 0 # 2 1720 0 0 -1 0 0 0 0 0 0 0 0 0 0 0 # 3 1730 0 0 0 0 0 -1 0 0 0 0 0 0 0 0 # 4 1740 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 # 5 1750 0 0 0 0 0 0 0 -1 0 0 0 0 0 0 # 6 1760 0 0 0 0 0 0 0 0 0 -1 0 0 0 0 # 7 1770 0 0 0 0 0 0 0 0 0 0 0 0 -1 0 # 8 1780 0 0 0 0 -1 0 0 0 0 0 0 0 0 0 # 9 1790 0 0 0 -1 0 0 0 0 0 0 0 0 0 0 # 10 1800 0 0 0 0 0 0 0 0 0 0 0 0 0 -1 # 11 1810 0 0 0 0 0 0 0 0 -1 0 0 0 0 0 # 12 182-1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 # 13 1830 0 0 0 0 0 0 0 0 0 0 0 -1 0 0 # 14 1840 0 0 -1 -1 -1 -1 0 0 -1 0 0 0 0 -1 # 15 185-1 -1 0 0 0 -1 0 0 0 0 0 0 -1 -1 -1 # 16 186-1 0 0 0 -1 0 0 0 -1 -1 -1 0 -1 0 0 # 17 1871 1 0 0 0 1 2 2 0 2 2 0 1 1 1 # 18 1881 0 2 2 1 0 0 0 1 1 1 0 1 2 2 # 19 1890 0 0 1 1 1 1 0 0 1 2 2 2 2 1 # 20 1901 1 0 2 2 1 0 2 2 0 0 0 1 1 1 # 21 1911 2 2 0 1 2 2 0 1 1 1 0 1 0 0 # 22 1922 2 0 1 1 1 1 0 2 1 0 2 0 0 1 # 23 1930 -1 0 -1 0 0 -1 0 -1 0 -1 0 0 -1 0 # 24 194 195N_RAYS 19625 197 198LINEALITY_SPACE 1990 0 0 0 0 1 1 1 1 1 1 1 1 1 1 2000 0 0 0 1 0 0 0 1 0 0 1 0 1 1 2010 0 0 1 0 0 0 1 0 0 1 0 1 0 1 2020 0 1 0 0 0 1 0 0 1 0 0 1 1 0 2030 1 0 0 0 0 -1 -1 -1 0 0 0 -1 -1 -1 2041 0 0 0 0 0 0 0 0 -1 -1 -1 -1 -1 -1 205 206ORTH_LINEALITY_SPACE 2070 0 0 0 0 0 0 0 0 0 1 -1 -1 1 0 2080 0 0 0 0 0 0 0 0 1 0 -1 -1 0 1 2090 0 0 0 0 0 0 1 -1 0 0 0 -1 1 0 2100 0 0 0 0 0 1 0 -1 0 0 0 -1 0 1 2110 0 0 0 0 1 0 0 -1 0 0 -1 -1 1 1 2120 0 0 1 -1 0 0 0 0 0 0 0 -1 1 0 2130 0 1 0 -1 0 0 0 0 0 0 0 -1 0 1 2140 1 0 0 -1 0 0 0 0 0 0 -1 -1 1 1 2151 0 0 0 -1 0 0 0 -1 0 0 0 -1 1 1 216 217F_VECTOR 2181 25 105 105 219\end{verbatim} 220After this follows a list of cones and maximal cones. 221Every maximal cone has an associated multiplicity which is also listed. 222 223The output says that the tropical variety has dimension $9$. Modulo the 224$6$-dimensional homogeneity space this is reduced to a $3$-dimensional complex in 225$\R^9$ and thus we may think of the tropical variety as a 226$2$-dimensional polyhedral complex on the $8$-sphere in $\R^9$. This 227complex is simplicial and has $105$ maximal cones. 228 229 230The extreme rays (modulo the homogeneity space) are labeled 231$0,\dots,24$. In the cone lists the cones are grouped together 232according to dimension and orbit with respect to the specified 233symmetries. See Section~\ref{sec:format fan} for more information on how to read the 234polyhedral fan format. 235\end{example} 236 237%If we look carefully in the debug output from the program we 238%will also see that these $105$ cones come in $17$ orbits. If we 239%wanted to know more about the traversed cones we could use the option 240%\texttt{--largedimensional}. This will list the $d$-dimensional and 241%$(d-1)$-dimensional cones of the fan. In our example one of the 242%$d$-dimensional cones is listed like this: (In this manual lines have 243%been concatenated to avoid wasting paper) 244 245While traversing the variety the program 246\texttt{gfan\_tropicaltraverse} only computes $d$ and 247$(d-1)$-dimensional cones. The other cones are extracted after 248traversing. Also the symmetries are expanded. Sometimes extracting all 249cones is time consuming and one is only interested in the high 250dimensional cones up to symmetry. These can be output using the option 251\texttt{--noincidence}. In that case the output would be a list of 252orbits for maximal cones and a list of orbits for codimension one 253cones. It is also listed how these cones are connected taking symmetry 254into account. In general that format is rather difficult to read. 255 256 257A final remark about \texttt{gfan\_tropicaltraverse} is that the 258polyhedral structure of the complex comes from the Gr\"obner fan. For 259some ideals it is possible to find polyhedral fans covering the 260tropical variety with fewer cones. 261 262 263\subsection{Intersecting tropical hypersurfaces} 264The tropical variety of a principal ideal is called a \emph{tropical 265hypersurface}. A \emph{tropical prevariety} is a finite intersection of 266tropical hypersurfaces or, to be precise, the intersection of the 267support set of these hypersurfaces. In Gfan the intersection is 268represented by the \emph{common refinement} of the tropical 269hypersurfaces. The program \texttt{gfan\_tropicalintersection} can 270compute such intersections. 271\begin{example} 272To compute the intersection of the tropical hypersurfaces $\T(\langle a+b+c+1\rangle)$ and $\T(\langle a+b+2c\rangle)$ we run 273\begin{verbatim} 274gfan_tropicalintersection 275\end{verbatim} 276on 277\begin{verbatim} 278Q[a,b,c] 279{a+b+c+1,a+b+2c} 280\end{verbatim} 281The output is a polyhedral fan whose support is the intersection. The 282balancing condition for this fan is not satisfied which implies that it 283is not a tropical variety. 284%list of cones whose union is the tropical prevariety. Notice, as a polyhedral complex, some cones might be missing from the list but all maximal cones are present. The program also gives a list with a relative interior point for each cone. 285 286%If we use the option \texttt{--incidence} we will get more information about the combinatorial structure of the intersection as a polyhedral complex. We should note that this option only investigates the maximal dimensional complex. 287 288%An interesting question is if the intersection equals the tropical variety of the ideal generated by the input polynomials. A necessary condition for this to be true is that all the computed relative interior points pick out monomial-free initial ideal. This can be checked with the option \texttt{-t}. In our example the prevariety is not equal to the tropical variety and the program will find a vector that proves this. 289\end{example} 290 291\subsection{Computing tropical bases of curves} 292In Gfan an ideal $I$ is said to define a \emph{tropical curve} if 293$k[\x_1,\dots,x_n]/I$ has Krull dimension equal to or one larger than the 294dimension of the homogeneity space of $I$. A \emph{tropical basis} of $I$ is 295a finite generating set for the ideal such that the tropical variety 296defined by $I$ (as a set) is the intersection of the tropical 297hypersurfaces of the generators. A tropical basis always exists \cite{ctv}. The 298program \textup{gfan\_tropicalbasis} computes a tropical basis for an 299ideal defining a tropical curve. 300\begin{example} 301Again we consider the ideal $\langle a+b+c+1,a+b+2c\rangle$. We notice that this ideal defines a curve since the Krull dimension is $1$ and the dimension of the homogeneity space is $0$. In the example above we saw that the listed set is not a tropical basis. We run 302\begin{verbatim} 303gfan_tropicalbasis -h 304\end{verbatim} 305on 306\begin{verbatim} 307Q[a,b,c] 308{a+b+c+1,a+b+2c} 309\end{verbatim} 310to get some tropical basis 311\begin{verbatim} 312Q[a,b,c] 313{ 314-1+c, 3152+b+a} 316\end{verbatim} 317We needed the option \texttt{-h} here since the ideal was not homogeneous. If we run \texttt{gfan\_tropicalintersection} on the output we see that the tropical variety consists of three rays and the origin. 318\end{example} 319 320 321\subsection{Tropical intersection theory} 322Gfan contains a few experimental programs for doing computations in 323tropical intersection theory. In \cite[Definition 3.4]{allermannRau} 324the tropical Weil divisor of a tropical rational function on a 325(tropical) $k$-cycle in $\R^n$ is defined. This divisor can be 326computed in Gfan. However, Gfan and \cite{allermannRau} do not agree 327on the basic definitions in tropical geometry. For example the 328definition of a fan is different. Here we will adjust the necessary 329definitions to the Gfan conventions. A tropical $k$-cycle will be a 330pure (rational) polyhedral fan $F$ of dimension $k$ in $\R^n$ with 331weights which is balanced in the following sense: To every 332$k$-dimensional facet $C$ we assign a weight (or multiplicity) 333$m_C\in\Z$. The vector space $\R^n$ comes with its standard lattice 334$\Z^n$. For a $k-1$-dimensional ridge $R\in F$ and a facet $C$ in its 335star\footnote{the smallest polyhedral subcomplex of $F$ containing all 336faces of $F$ containing $R$.} in $F$ corresponding to a cone $L$ in 337the link\footnote{take an $\epsilon$-ball around a relative interior 338$\omega\in R$ and intersect it with $F$. Translating the ball to the 339origin and scaling the intersection to infinity we get the link of $R$ 340in $F$.} of $R$ in $F$, the semi-group 341$L\cap\Z^n/\textup{span}_{\R}(R)\cap\Z^n\subseteq 342\Z^n/\textup{span}_{\R}(R)\cap\Z^n$ is isomorphic to $\N$. Define 343$u_{C/R}\in \Z^n/\textup{span}(R)\cap\Z^n$ as the element identified 344with $1\in\N$. The balancing condition at $R$ is that 345$$\sum_{C\in F:R\subset C} m_Cu_{C/R}=0.$$ 346For a (weighted) fan to be a tropical cycle this must hold at every ridge $R$. 347 348It remains to define what a tropical rational function is. Take a 349polyhedral fan $F'$ and associate to each of its maximal cones a 350linear form. When evaluating a point $x$ in the support of $F'$ simply 351evaluate the linear form of cone containing $x$. If this gives a 352well-defined function we call this function a tropical rational 353function. When computing Weil divisors we will require that the supports satisfy 354$\textup{supp}(F)\subseteq \textup{supp}(F')$. There will be no further restriction 355on the polyhedral structure. 356 357For a definition of the Weil divisor itself we refer 358to \cite[Definition 3.4]{allermannRau}. Here we just mention that it 359again is a cycle of dimension one lower. 360 361To demonstrate the Gfan features we recompute \cite[Example 3.10]{allermannRau}. 362An easy way to generate the $k$-cycle of that example is to compute it as a hypersurface. Since the paper is using min and Gfan is using max we need to change the polynomial from the paper such that the Newton polytope is flipped: 363\begin{verbatim} 364gfan_tropicalhypersurface > tmpfile1.poly 365Q[x_1,x_2,x_3] 366{x_2x_3+x_1x_3+x_1x_2+x_1x_2x_3} 367\end{verbatim} 368The weights/multiplicities are stored in the MULTIPLICITIES section of the Polymake file. 369 370It is harder specifying the rational function. We make the following file and call in \texttt{func.poly}. 371\begin{footnotesize} 372\begin{verbatim} 373_application PolyhedralFan 374_version 2.2 375_type PolyhedralFan 376 377AMBIENT_DIM 3783 379 380DIM 3812 382 383LINEALITY_DIM 3840 385 386RAYS 3871 0 0 # 0 3880 1 0 # 1 3890 0 1 # 2 390-1 -1 -1 # 3 3911 1 0 # 4 392-1 -1 0 # 5 393 394N_RAYS 3956 396 397LINEALITY_SPACE 398 399ORTH_LINEALITY_SPACE 4001 0 0 4010 1 0 4020 0 1 403 404MAXIMAL_CONES 405{3 5} # Dimension 2 406{5 2} 407{0 2} 408{1 2} 409{1 3} 410{0 3} 411{1 4} 412{0 4} 413 414MULTIPLICITIES 4151 4161 4171 4181 4191 4201 4211 4221 423 424RAY_VALUES 4250 4260 4270 4281 429-1 4300 431 432LINEALITY_VALUES 433\end{verbatim} 434\end{footnotesize} 435Instead of specifying the linear function on each maximal cone we have 436to specify its values on each of the rays in the fan and each of the 437generators of the lineality space. Then Gfan will automatically 438interpolate the function. Since the lineality space of the fan is 439empty we leave the LINEALITY\_VALUES section empty. 440 441We now compute the Weil divisor: 442\begin{footnotesize} 443\begin{verbatim} 444gfan_tropicalweildivisor -i1 tmpfile1.poly -i2 func.poly >tmpfile2.poly 445\end{verbatim} 446\end{footnotesize} 447...and compute the Weil divisor again as in \cite{allermannRau}... 448\begin{footnotesize} 449\begin{verbatim} 450gfan_tropicalweildivisor -i1 tmpfile2.poly -i2 func.poly >tmpfile3.poly 451\end{verbatim} 452\end{footnotesize} 453We get a fan with the origin being the only cone. It has multiplicity $-1$: 454\begin{verbatim} 455MULTIPLICITIES 456-1 # Dimension 0 457\end{verbatim} 458 459There is another useful command for computing polyhedral fans for 460rational functions. The command \texttt{gfan\_tropicalfunction} takes a 461polynomial and turns it into a fan representing its tropicalization 462which is a tropical rational function. 463 464\subsection{Non-constant coefficients} 465\label{sec:non-constant} 466In tropical geometry it is common to take the valuation of 467$\CC\{\{t\}\}$ into account when defining the tropical variety of 468an ideal in $\CC\{\{t\}\}[x_1,\dots,x_n]$. Here $\CC\{\{t\}\}$ denotes the field of 469Puiseux series. The valuation $\textup{val}(p)$ of a non-zero Puiseux 470series $p$ is the degree of its lowest order term. 471 472 473\begin{definition} 474For $\omega\in\R^n$ the \emph{t-$\omega$-degree}\index{t-$\omega$-degree} of a term $ct^ax^v$ 475with $c\in\CC^*$, $a\in \Q$ and $v\in\Z^n$ is defined as 476$-\val(ct^a)+\omega\cdot v=-a+\omega\cdot v$. The \emph{t-initial 477form}\index{t-initial form} $\tinit_\omega(f)\in\CC[x_1,\dots,x_n]$ of a polynomial 478$f\in\puiseux[x_1,\dots,x_n]$ is the sum of all terms in $f$ of maximal 479t-$\omega$-weight but with $1$ substituted for $t$. 480\end{definition} 481\begin{remark} 482Notice that since $t$ has t-$\omega$-degree $-1$, the maximal 483t-$\omega$-weight \emph{is} attained by a term if the polynomial is 484non-zero. Furthermore, only a finite number of terms attain the 485maximum. Therefore, it makes sense to substitute $t=1$ and consider 486the finite sum of terms as a polynomial in $\CC[x_1,\dots,x_n]$. 487\end{remark} 488\begin{example} 489Consider $f=(1+t)+t^2x+tx^2\in\puiseux[x_1,\dots,x_n]$. Let $\omega=({1\over 4902})\in\R^1$. Then $\tinit_\omega(f)=1+x^2$. For any 491other choice of $\omega$ the t-initial form is a monomial. 492\end{example} 493\begin{definition} 494Let $I\subseteq \puiseux[x_1,\dots,x_n]$ and $\omega\in\R^n$. The \emph{t-initial ideal}\index{t-initial ideal} of $I$ with respect to $\omega$ is defined as: 495$$\tinit_\omega(I):=\langle \tinit_\omega(f):f\in I\rangle\subseteq\CC[x_1,\dots,x_n].$$ 496\end{definition} 497 498\begin{definition} 499\label{def:tropvar} 500Let $I\subseteq \puiseux[x_1,\dots,x_n]$ be an ideal. The \emph{tropical variety} of $I$ is the set 501$$\T'(I):=\{\omega\in\R^n:\tinit_\omega(I) \textup{ is monomial-free}\}.$$ 502%Here monomial-free\index{monomial-free} means that the ideal does not contain a monomial. 503\end{definition} 504We use the notation $\T'(I)$ to avoid contradicting our original definition 505of the tropical variety of an ideal in the polynomial ring over a 506field. 507%An important theorem says that the tropical variety of $I$ is also the 508%negative of the closure of the image of $V(I)\subseteq \CC\{\{t\}\}^*$ 509%under the coordinatewise valuation. 510 511 512\begin{proposition} \cite[Proposition~7.3]{lifting} 513\label{prop:computing tinit} 514Let $I\subseteq \CC[t,x_1,\dots,x_n]$ be an ideal, $J=\langle I\rangle_{\puiseux[x_1,\dots,x_n]}$ and $\omega\in\R^n$. Then $\textup{t-in}_\omega(I)=\textup{t-in}_\omega(J)$. 515\end{proposition} 516 517\begin{remark} 518\label{rem:computing tinit} 519For $f\in\CC[t,x_1,\dots,x_n]$ we have 520$\tinit_\omega(f)=(\init_{(-1,\omega)}(f))|_{t=1}$. Consequently, for 521$I\subseteq\CC[t,x_1,\dots,x_n]$ we have 522$\tinit_\omega(I)=(\init_{(-1,\omega)}(I))|_{t=1}$. In order to 523decide if $\tinit_\omega(I)$ contains a monomial we may simply decide if the initial ideal 524$\init_{(-1,\omega)}(I)$ contains a monomial. 525 As a corollary we get 526$$\T(I)\cap(\{-1\}\times\R^n)=\{-1\}\times\T'(J).$$ 527 528In fact this gives a method for computing the tropical variety as a set of any 529ideal $J\subseteq\CC\{\{t\}\}[x_1,\dots,x_n]$ generated by elements 530in the polynomial ring over the field of rational functions 531$\Q(t)[x_1,\dots,x_n]$ in Gfan by clearing denominators and 532intersecting the result with the $t=-1$ plane. (We remind the reader 533that Lemma~\ref{lem:tropical by homogenisation} shows that for 534computational purposes it is no restriction if $I$ is not 535homogeneous.) 536\end{remark} 537 538Intersecting the tropical variety with the $t=-1$ plane can with some difficulty be done by 539hand. If the tropical (pre)-variety has been computed with 540\texttt{gfan\_tropicalintersection} then it is also possible to let Gfan do 541the intersection. What Gfan does is to compute the common refinement 542of the fan with the fan consisting of the halfspace $t\leq 0$ and its 543proper face. Of course this does not remove the cones in the $t=0$ 544plane, but they are easily removed by hand. We illustrate the 545procedure by an example. 546 547\begin{example} 548\label{ex:nonconstant} 549Exercise 2 in Chapter 9 of \cite{sturmfelssolving} asks us to draw the variety 550defined by the \emph{tropical} polynomial 551$f=1x^2+2xy+1y^2+3x+3y+1$. If we tropically divide this polynomial by $3$ we get $f':=f/3=-2x^2-1xy-2y^2+0x+0y+-2$ which defines the same tropical variety. This variety equals the variety defined by 552the polynomial $g=t^2x^2+txy+t^2y^2+x+y+t^2\in\CC\{\{t\}\}[x,y]$. Notice that $f'$ is the tropicalisation of $g$. 553 554According to Remark~\ref{rem:computing tinit} above the we may compute $\T'(\langle g\rangle)$ by computing 555the variety of $\langle t^2x^2+txy+t^2y^2+x+y+t^2\rangle\subseteq \CC[t,x,y]$ and intersecting it with the hyperplane $t=-1$. 556Running 557\begin{verbatim} 558gfan_tropicalintersection --tplane 559\end{verbatim} 560on 561\begin{verbatim} 562Q[t,x,y] 563{t^2x^2+txy+t^2y^2+x+y+t^2} 564\end{verbatim} 565we get 566\begin{verbatim} 567RAYS 5680 -1 0 # 0 569-1 2 1 # 1 5700 1 1 # 2 571-1 1 1 # 3 572-1 -2 -2 # 4 5730 0 -1 # 5 574-1 1 2 # 6 575 576MAXIMAL_CONES 577{3 4} # Dimension 2 578{2 6} 579{1 3} 580{1 2} 581{3 6} 582{4 5} 583{0 4} 584{0 6} 585{1 5} 586\end{verbatim} 587among other information. We can now draw the two-dimensional picture 588asked for in the exercise. The rays with non-zero first coordinate 589become points in the picture. (If the first coordinate is not $-1$ scaling 590is required to get the rational $x,y$-coordinates.) The rays with zero 591first coordinate become directions. The maximal cones show how to 592connect the rays; see Figure~\ref{fig:nonconstant}. Notice that some 593of the connections could have been ``at infinity''. 594\begin{figure} 595\begin{center} 596\epsfig{file=nonconst.eps,height=5.5cm} 597\end{center} 598\caption{The tropical variety defined by the tropical polynomial in Example~\ref{ex:nonconstant}.} 599\label{fig:nonconstant} 600\end{figure} 601 602\end{example} 603 604\subsubsection{Algebraic field extensions of $\Q$} 605Ignoring time, memory usage and overflows Gfan can compute the tropical variety $\T'(I)$ of any ideal $I\subseteq \puiseux[x_1,\dots,x_n]$ generated by elements of $\overline{\Q}(t)[x_1,\dots,x_n]$. This is a consequence of the following lemma: 606\begin{lemma}\cite[Lemma~3.12]{lifting} 607\label{lem:fieldextension} 608Let $k$ be a field and $M=\langle m\rangle\subseteq k[a]$ a maximal ideal where $m$ is not a monomial. Let 609$I\subseteq (k[a]/M)[x_1,\dots,x_n]$ be an ideal. For $\omega\in\R^n$ we 610have 611$$\init_\omega(I) \textup{ contains a monomial} \Longleftrightarrow \init_{(0,\omega)}(\varphi^{-1}(I)) \textup{ contains a monomial}$$ 612where $\varphi:k[a,x_1,\dots,x_n]\rightarrow (k[a]/M)[x_1,\dots,x_n]$ is the homomorphism taking elements to their cosets. 613\end{lemma} 614