A Two-Dimensional Improvement for Farr-Gao Algorithm

Assume that \Xi is \mathcal{A}-Cartesian satisfying (2.1) where \mathcal{A} can be represented as (?) that together with (2.1) implies \mathrm{b}_y(\mathcal{A})=\#\pi_y(\Xi)-1 and

Since m_0\geq m_1\geq \cdots\geq m_{b_y(\mathcal{A})}\geq 0, (2.2) follows. (2.2) can be proved in like manner.

Conversely, we suppose that (2.2) holds and that lower sets S_x(\Xi) and S_y(\Xi) have the expression (?). Then there exists a unique sequence (n_0, n_1, \ldots,n_{\#\Xi_0^x-1})\in \mathbb{N}_0^{\#\Xi_0^x} such that

where n_0=\#\pi_y(\Xi)-1.

Next, we shall verify that \#\pi_x(\Xi)=\#\Xi_0^x, namely we can cover \Xi by exactly \#\Xi_0^x vertical lines. Prove this by contradiction. It is evident from \#\pi_x(\Xi)\geq \#\Xi_0^x that the equality fails only when \#\pi_x(\Xi)> \#\Xi_0^x, i.e., there exists at least one point (\bar{x}, \bar{y})\in \Xi such that \bar{x}\notin \pi_x(\Xi_0^x). However, (\bar{x}, \bar{y})\in \Xi simply means that there exists some 0<j\leq \#\pi_y(\Xi)-1 such that (\bar{x}, \bar{y})\in \Xi_j^x. Thus (2.2) implies that \bar{x}\in \pi_x(\Xi_j^x)\subseteq \pi_x(\Xi_0^x) which contradicts \bar{x}\notin \pi_x(\Xi_0^x).

In the rest of the proof, we will use induction on h to show that

which leads to S_x(\Xi)=S_y(\Xi) immediately. When h=0, for every (\bar{x}, \bar{y})\in \Xi_{\#\pi_y(\Xi)-1}^x, it follows from (2.2) that \bar{x}\in \pi_x(\Xi_{\#\pi_y(\Xi)-1}^x)\subset \pi_x(\Xi_j^x), 0\leq j\leq \#\pi_y(\Xi)-2, which means that \#\{\Xi\cap \{x=\bar{x}\}\}\geq \#\pi_y(\Xi). But \#\{\Xi\cap \{x=\bar{x}\}\}\leq \#\pi_y(\Xi) is trivial, we have \#\Xi_0^y=\#\{\Xi\cap \{x=\bar{x}\}\}=\#\pi_y(\Xi)=n_0+1, namely (?) is true for h=0.

Now assume (?) for 0\leq h \leq k< \#\pi_x(\Xi)-1, i.e., n_h=\#\Xi_h^y-1, h=0, \ldots, k. It turns out that there exist distinct \bar{x}_0, \bar{x}_1, \ldots, \bar{x}_k\in \pi_x(\Xi) such that \#\{\Xi\cap\{x=\bar{x}_h\}\}=n_h+1, 0\leq h \leq k. Thus, for every 0\leq h \leq k, n_h\geq n_{k+1} implies \bar{x}_h\in \pi_x(\Xi_{n_{k+1}}^x). Since (k+1, n_{k+1})\in S_x(\Xi), there exists at least one point (\bar{x}_{k+1}, \bar{y}_{k+1})\in \Xi_{n_{k+1}}^x that is not in \Xi_j^y, j=0, \ldots, k. By (2.2), a similar argument leads to \#\{\Xi\cap \{x=\bar{x}_{k+1}\}\}\geq n_{k+1}+1 which implies \#\Xi_{k+1}^y\geq n_{k+1}+1. On the other side, it follows from induction hypothesis that every point in \Xi_{k+1}^y belongs to some \Xi_j^x, 0\leq j\leq n_{k+1}, which implies that \#\Xi_{k+1}^y\leq n_{k+1}+1, therefore we have \#\Xi_{k+1}^y=n_{k+1}+1, namely (?) holds for h=k+1. Theorem 2.1 immediately implies that \Xi is Cartesian.

Swapping the roles of x and y, the other statement can be proved similarly.

We only offer the proof of the first statement. The second one can be verified in very like fashion.

Retain all the notation established previously. Fix monomial order as \prec_{\mathrm{grlex}}, and for simplicity the symbol will be omitted in the rest of the proof where no confusion arises. Suppose that x-tower set \Xi has the decompositions (?). For convenience, we set m_j:=\#\Xi_j^x-1, j=0, \ldots, \nu, where \nu=\#\pi_y(\Xi)-1=\mathrm{b}_y(S_x(\Xi)). Thus, (?) implies S_x(\Xi)=\mathrm{L}_x(m_0, m_1, \ldots, m_\nu). Fix 0\leq j\leq \nu. By Lemma 3.1, the ideal

Obviously, \mathcal{I}_j’s are pairwise comaximal. Hence

Let G_N be the reduced Gröbner basis for ideal \prod_{j=0}^N \mathcal{I}_j w.r.t. \prec_{\mathrm{grlex}}, 0\leq N \leq \nu. We will use induction on N to prove

First of all,

leads to (?) immediately for N=0.

It follows from Theorem 2.4 that m_0 > m_1 and \pi_x(\Xi_1^x)\subsetneq \pi_x(\Xi_0^x). Therefore, we have

where the last equality holds by

Then \mathrm{LM}(G_1) = \{x^{m_0+1},x^{m_1+1}y,y^2\} follows which means that (?) holds for N=1.

Similarly, by m_1>m_2 and \pi_x(\Xi_2^x)\subsetneq \pi_x(\Xi_0^x), we obtain, after some easy computations,

We let \hat{q}, \hat{r}\in \Pi^1 be the quotient and the remainder of the division of \prod_{(i, 1)\in S_x(\Xi)} (x-\mathrm{x}(\sigma_1^x(i))) by \prod_{(i, 2)\in S_x(\Xi)} (x-\mathrm{x}(\sigma_2^x(i))) respectively, namely

Denote the remainder of g_3^{(2)} w.r.t. g_4^{(2)} by \bar{g}_3^{(2)}. One can check readily that

On the other hand,

implies g_2^{(2)} \xrightarrow{\{g_3^{(2)}, g_4^{(2)}\}}_+ 0. Consequently,

We claim that

where \mathrm{y}(1)-\mathrm{y}(2)=\mathrm{LC}\left(\bar{g}_3^{(2)}\right), i.e., (\mathrm{y}(1)-\mathrm{y}(2))^{-1}\bar{g}_3^{(2)} is monic.

In fact, if S(f, g) stands for the S-polynomial of polynomials f, g\in \Pi^2, then S\left(g_0^{(2)}, g_1^{(2)}\right)\xrightarrow{G'_2}_+ 0 follows immediately because \mathrm{LM}\left(g_0^{(2)}\right) and \mathrm{LM}\left(g_1^{(2)}\right) are relatively prime. Observing that \prod_{(i, 2)\in S_x(\Xi)} (x-\mathrm{x}(\sigma_2^x(i))) is a factor of \prod_{(i, 0)\in S_x(\Xi)} (x-\mathrm{x}(\sigma_0^x(i))), we get

It is easy to see that \mathrm{LM}\left(S\left(g_0^{(2)}, g_4^{(2)}\right)\right)=x^{m_0+1}y. Moreover, a simple computation leads to:

therefore

implies S\left(g_0^{(2)}, g_4^{(2)}\right)\xrightarrow{G_2'}_+ 0. Similarly,

and \mathrm{LM}\left(S\left(g_1^{(2)}, g_4^{(2)}\right)\right)=x^{m_2+1}y^2 imply

which means that S\left(g_1^{(2)}, g_4^{(2)}\right)\xrightarrow{G_2'}_+ 0.

In like manner, we can also show that

Hence, there only remains S\left((\mathrm{y}(1)-\mathrm{y}(2))^{-1}\bar{g}_3^{(2)}, g_4^{(2)}\right) \xrightarrow{G_2'}_+ 0 to be verified. Actually, it can be deduced that

whose leading monomial is x^{m_1+1}y. Thus

and

imply that S\left((\mathrm{y}(1)-\mathrm{y}(2))^{-1}\bar{g}_3^{(2)}, g_4^{(2)}\right) \xrightarrow{G_2'}_+ 0.

Now, these arguments lead to the conclusion that S-polynomial S(g, g')\xrightarrow{G'_2}_+ 0 for any two distinct g, g' \in G'_2. By Buchberger’s S-pair criterion, G_2' is a Gröbner basis for \mathcal{I}_0 \cdot \mathcal{I}_1 \cdot \mathcal{I}_2 w.r.t. \prec_{\mathrm{grlex}}. Moreover, for every g\in G_2', it is evident that

1. \mathrm{LC}(g)=1,

2. No monomial of g lies in \langle \mathrm{LT}(G_2'-\{g\})\rangle,

which means that G_2' is reduced, namely (?) holds for N=2.

Now, assume (?) for N=k, 3\leq k<\nu.

Without loss of generality, we suppose that G_k = \left\{g_0^{(k)}, \ldots, g_{k+1}^{(k)}\right\} with

which imply that

When N=k+1, since \Xi is x-tower, we obtain

where g_0^{(k+1)}=g_0^{(k)} is obvious hence g_0^{(k)}(y-\mathrm{y}(k+1)) can be removed. By the induction hypothesis, we have

We denote polynomial g_{k+1}^{(k)}(\prod_{(i, k+1)\in S_x(\Xi)} (x-\mathrm{x}(\sigma_{k+1}^x(i)))) by g_{k+1}^{(k+1)}. It follows from the induction hypothesis that \mathrm{LM}\left(g_{k+1}^{(k+1)}\right)=x^{m_{k+1}+1}y^{k+1}. Set E_1:=\left\{g_{k+1}^{(k+1)}\right\}. Suppose that g_k^{(k)}(y-\mathrm{y}(k+1))\xrightarrow{E_1}_+ g_k^{(k+1)}. Since m_k > m_{k+1}, we have

Recall case N=2. It is easy to see that

where F_1:=\left\{g_{k+1}^{(k+1)}, g_k^{(k+1)}\right\}, which means that g_k^{(k)}(\prod_{(i, k+1)\in S_x(\Xi)} (x-\mathrm{x}(\sigma_{k+1}^x(i)))) can be removed from the original ideal basis.

Set E_2:=F_1, and suppose that g_{k-1}^{(k)}(y-\mathrm{y}(k+1)) \xrightarrow{E_2}_+ g_{k-1}^{(k+1)}. We similarly deduce that \mathrm{LM}\left(g_{k-1}^{(k+1)}\right)=x^{m_{k-1}+1}y^{k-1} and

where F_2:=\left\{g_{k+1}^{(k+1)}, g_{k-1}^{(k+1)}\right\}.

In this way, we construct two sequences (E_1, E_2,\ldots, E_k) and (F_1, F_2, \ldots, F_k) such that

where, for i=1,\ldots,k,

Let \bar{g}_i^{(k+1)}=\mathrm{LC}\left(g_i^{(k+1)}\right)^{-1} g_i^{(k+1)}, i=0, \ldots, k+2. With similar arguments used in N=2 case, we can finally prove that

with \mathrm{LM}\left(\bar{g}_i^{(k+1)}\right)=x^{m_i+1}y^i, i=0, \ldots, k+1, and \mathrm{LM}\left(\bar{g}_{k+2}^{(k+1)}\right)=y^{k+2}, namely (?) holds for N=k+1. Consequently, we have

Recalling (?) and (?), we have

which complete the proof.

Fix (i, j)\in S_x(\Xi). If (i, j)=(m, n), then

Otherwise, (i, j)\succneqq_{\mathrm{inlex}} (m, n) implies j>n or j=n and i>m. When j>n, i.e., j-1\geq n, we have

If j=n, i>m, namely i-1\geq m, then

which leads to

as desired. It is easy to check that \mathrm{Span}_\mathbb{F}\mathfrak{N}_x=\mathrm{Span}_\mathbb{F}\mathfrak{M}_{S_x(\Xi)}=\mathcal{P}_{\prec_{\mathrm{grlex}}}(\Xi).