betaFeedback

WorksMathematics

A counterexample to the Eisenbud–Mazur conjecture

Abstract

We construct a height-five prime ideal PP in the six-variable formal power series ring RR over Q\mathbb{Q} such that P(2)⊈mPP^{(2)}\not\subseteq\mathfrak{m}P, thus providing a prime-ideal counterexample to the Eisenbud–Mazur conjecture (with Q\mathbb{Q} being the coefficient field). The ideal is the kernel of an explicit quadratic parametrization over a number field of degree 2828 over Q\mathbb{Q}. A witness has order three, with leading term the determinant of a symmetric 3×33\times3 matrix. We construct the witness by eliminating critical values to all orders and then straightening the resulting formal curve. Interpolation at the 2828 nodes of an arrangement of eight lines supplies the required lifting maps. A rational linear functional on third and fourth jets certifies nonmembership in mP\mathfrak{m}P. The construction also yields an unmixed radical ideal with 2828 smooth branches over C\mathbb{C}, and provides a counterexample to the more general form of the Eisenbud–Mazur conjecture stated by More (with C\mathbb{C} being the coefficient field). Finally, the prime ideal counterexample descends to the localized polynomial ring Q[x0,…,x5](x0,…,x5)\mathbb{Q}[x_0,\ldots,x_5]_{(x_0,\ldots,x_5)}, where a polynomial witness with the same fourth jet exists. All finite identities used in the proof admit exact verification.

Introduction

The Eisenbud–Mazur conjecture asks whether

P(2)⊆mP(1)P^{(2)} \subseteq\mathfrak{m}P \tag*{(1)}

for every prime ideal PP in a regular local ring (R,m)(R,\mathfrak{m}) containing a field of characteristic zero. Here P(2)=P2RP∩RP^{(2)} = P^2R_P \cap R is the symbolic square of PP. Equivalently, it predicts that no element of the symbolic square can be a minimal generator of PP. Eisenbud and Mazur related this containment to the existence of nontrivial evolutions of reduced local algebras [1]. The formulation for arbitrary regular local rings containing a characteristic-zero field, and more generally for unmixed radical ideals, appears in [4 Conjecture 1.5].

We construct a counterexample to the original Eisenbud–Mazur conjecture in the formal power series ring in six variables over Q\mathbb{Q}. The quotient by PP is a one-dimensional complete local domain, and its normalization has a residue-field extension of Q\mathbb{Q} of degree 28. After extending coefficients to C\mathbb{C}, the construction gives a counterexample to [4 Conjecture 1.5] with 28 smooth branches. We also give a counterexample to the original Eisenbud–Mazur conjecture in a localized polynomial ring over Q\mathbb{Q}; this is established in Theorem 8.1.

There are substantial positive results for special classes of ideals. Eisenbud and Mazur prove the containment for generically complete intersection licci ideals, including perfect ideals of height two and Gorenstein ideals of height three, and for generically complete intersection almost complete intersections [1]. Huneke and Ribbe obtain further positive results using residual intersections, integral closure, and differentials [3]. More proves the conjecture for prime ideals obtained by contracting a parametrized smooth curve along the extension k[[t2,x1,…,xm]]⊂k[[t,x1,…,xm]]k[[t^2,x_1,\ldots,x_m]] \subset k[[t,x_1,\ldots,x_m]] [4 Theorem 2.1]. A related uniform bound is P(h+1)⊆mPP^{(h+1)} \subseteq\mathfrak{m}P, where hh is the height of PP: Hochster and Huneke prove this for regular local rings containing a field [2 Theorems 3.5 and 4.2], and Murayama proves a generalization for arbitrary regular local rings [5 Theorem C]. For the height-five prime ideal constructed below, this bound concerns the sixth symbolic power and is compatible with the failure of (1).

Two features of the problem account for both these positive results and the difficulty of finding a counterexample. First, in characteristic zero the symbolic square of a radical ideal has a differential description: f∈P(2)f \in P^{(2)} if and only if ff and all its partial derivatives lie in PP [1 Theorem 3]. So (1) is a statement about critical loci: a witness is a hypersurface f=0f=0 that is singular along V(P)V(P) and is nevertheless a minimal generator of PP. Second, if PP is stable under a derivation ∑wixi∂i\sum w_i x_i \partial_i with positive integer weights, the Euler identity ∑wixi∂if=(deg⁡f)f\sum w_i x_i \partial_i f=(\deg f)f places every such ff in mP\mathfrak{m}P [1 Proposition 10]. A counterexample must therefore be non-quasi-homogeneous in every system of coordinates. The hypothesis on the characteristic is essential as well: in characteristic pp the differential description breaks down, and [1] exhibits monomial curves in characteristics 2 and 3 that violate (1).

We set

R=Q[[x0,x1,x2,x3,x4,x5]],m=(x0,x1,x2,x3,x4,x5).R=\mathbb{Q}[[x_0,x_1,x_2,x_3,x_4,x_5]],\qquad\mathfrak{m}=(x_0,x_1,x_2,x_3,x_4,x_5).

The first two nonzero homogeneous terms of our witness are the cubic

D=det⁡(x0x1x2x1x3x4x2x4x5)=x0x3x5−x0x42−x12x5+2x1x2x4−x22x3(2)\begin{aligned} D&=\det\begin{pmatrix} x_0 & x_1 & x_2\\ x_1 & x_3 & x_4\\ x_2 & x_4 & x_5 \end{pmatrix}\\ &=x_0x_3x_5-x_0x_4^2-x_1^2x_5+2x_1x_2x_4-x_2^2x_3 \tag*{(2)} \end{aligned}

and the quartic

W=x04+x03x1−x03x2−2x02x32+16x0x1x3x5−7x0x2x32+14x0x3x4x5−20x0x3x52+8x1x53−7x2x3x52+x34+x4x53−x54.(3)W = x_0^4 + x_0^3x_1 - x_0^3x_2 - 2x_0^2x_3^2 + 16x_0x_1x_3x_5 - 7x_0x_2x_3^2 + 14x_0x_3x_4x_5 - 20x_0x_3x_5^2 + 8x_1x_5^3 - 7x_2x_3x_5^2 + x_3^4 + x_4x_5^3 - x_5^4. \tag*{(3)}

The significance of the quartic WW is explained after Theorem 1.1; see also § 3 and § 4.

Theorem 1.1. There is an explicitly defined continuous Q\mathbb{Q}-algebra homomorphism

φ:R⟶L[[t]],φ(x0)=t,φ(xi)=tℓi+t2wi(1≤i≤5),\varphi: R \longrightarrow L[[t]], \qquad\varphi(x_0) = t, \qquad\varphi(x_i) = t\ell_i + t^2w_i \quad(1 \le i \le5),

where LL is a number field of degree 28 over Q\mathbb{Q}, with ℓi,wi∈L\ell_i,w_i \in L for 1≤i≤51 \le i \le5, such that P=ker⁡φP = \ker\varphi is a height-five prime ideal and there exists f∈Rf \in R satisfying

f≡D+W(modm5),f∈P(2),f∉mP.(4)f \equiv D + W \pmod{\mathfrak{m}^5}, \qquad f \in P^{(2)}, \qquad f \notin\mathfrak{m}P. \tag*{(4)}

The normalization of R/PR/P is L[[t]]L[[t]], whereas the residue field of R/PR/P is Q\mathbb{Q}.

The field LL and all coefficients of φ\varphi are specified by rational polynomial formulas below. The witness ff in Theorem 1.1 is a complete formal power series, given in equation (30).

The construction combines a geometric lifting argument with a finite obstruction to membership in mP\mathfrak{m}P. The projectivized singular locus of DD is the quadratic Veronese surface: the partial derivatives of DD are the 2×22 \times2 minors, which generate the prime ideal QQ of the cone over this surface, so that D∈Q(2)D \in Q^{(2)}. The Laplace expansion, however, writes DD as a combination of the minors with coefficients in m\mathfrak{m}, so D∈mQD \in\mathfrak{m}Q and QQ itself is no counterexample. We therefore replace QQ by the ideal of a formal arc whose tangent line lies on the cone but which leaves the cone at second order. Such an ideal no longer contains all the minors, so the Laplace expansion no longer applies, and an element of order three with leading form DD can be a minimal generator; what has to be arranged is that its gradient vanishes along the arc, and the quartic WW supplies the critical point that makes this possible. On its affine chart

(x0,x1,x2,x3,x4,x5)=(1,U,V,U2,UV,V2),(x_0,x_1,x_2,x_3,x_4,x_5) = (1,U,V,U^2,UV,V^2),

the restriction of WW is the product of the eight lines U+αV+α2=0U+\alpha V+\alpha^2=0, where α\alpha runs through the roots of z8−z−1z^8-z-1. Their 28 nodes form a single Galois orbit. Evaluation at one node identifies the polynomials in U,VU,V of degree at most six with LL. Consequently, evaluation at the corresponding point of the Veronese surface maps homogeneous cubics in the six variables onto LL, and the same holds in every higher degree.

The interpolation has two uses. First, the cubic determinant has a nondegenerate quadratic form in the directions normal to the Veronese surface, and the restricted quartic has a nondegenerate critical point at each node. Two successive applications of the formal implicit function theorem therefore produce a critical arc. A quintic term removes the first obstruction to its critical value being zero; the surjectivity of evaluation then removes all subsequent obstructions, one homogeneous degree at a time. Second, the same interpolation gives φ(m3)=t3L[[t]]\varphi(\mathfrak{m}^{3})=t^{3}L[[t]]. This permits a formal coordinate change with identity two-jet that replaces the critical arc by the prescribed quadratic parametrization while preserving D+WD+W modulo m5\mathfrak{m}^{5}. The two-term shape of the arc and of the witness is what defeats the Euler argument: no positive grading makes both homogeneous.

The obstruction to f∈mPf\in\mathfrak{m}P can be checked using only the fourth jet. We determine the image of PP modulo m4\mathfrak{m}^{4} and display a rational linear functional supported in degrees three and four after quotienting by x0x_{0}. It annihilates the image of mP\mathfrak{m}P and takes the value 1 on D+WD+W. Note that the proof of symbolic-square membership uses the full formal series, whereas the nonmembership argument depends only on its fourth jet. The exact computations needed to verify the finite identities are recorded in the appendix.

Finally, Proposition 7.1 describes the extension of PP to C[[x0,…,x5]]\mathbb{C}[[x_{0},\ldots,x_{5}]] as the intersection of 28 distinct smooth branch primes. This also gives a counterexample to the extended form of the Eisenbud–Mazur conjecture considered in [4 Conjecture 1.5].

Declaration of generative AI: The counterexample was discovered by a ChatGPT model. The authors reviewed and edited the content as needed and take full responsibility for the content of the published article.

The coefficient field LL

Throughout the paper, orders and homogeneous components refer to ordinary total degree in x0,…,x5x_{0},\ldots,x_{5}. We write Hd=Q[x0,…,x5]dH_{d}=\mathbb{Q}[x_{0},\ldots,x_{5}]_{d} for the space of homogeneous forms of degree dd. Let

h(z)=z8−z−1,h(z)=z^{8}-z-1,

choose two distinct roots β,γ\beta,\gamma in an algebraic closure of Q\mathbb{Q}, and put

u=βγ,v=−(β+γ),L=Q(u,v).(5)u=\beta\gamma,\qquad v=-(\beta+\gamma),\qquad L=\mathbb{Q}(u,v). \tag*{(5)}

Lemma 2.1. The extension L/QL/\mathbb{Q} has degree 28. Its embeddings into a splitting field of hh are indexed by the unordered pairs of distinct roots of hh.

Proof. The polynomial hh is irreducible modulo 7. Modulo 19 it factors as

h(z)=(z−4)(z7+4z6−3z5+7z4+9z3−2z2−8z+5),(6)h(z)=(z-4)\left(z^{7}+4z^{6}-3z^{5}+7z^{4}+9z^{3}-2z^{2}-8z+5\right), \tag*{(6)}

where the degree-seven factor is irreducible. Both reductions are squarefree; exact certificates are given in Appendix A. Consequently the Galois group of hh is transitive on its eight roots and contains a 7-cycle. The stabilizer of the root fixed by this cycle acts transitively on the other seven roots. Transitivity of the whole group therefore implies double transitivity, and hence transitivity on the 28 unordered pairs of distinct roots.

The pair (u,v)(u,v) determines {β,γ}\{\beta,\gamma\} as the roots of z2+vz+uz^2+vz+u. Its orbit thus has size 28, which is [L:Q][L:\mathbb{Q}]. □

For each root α\alpha of hh, consider the line

Λα:U+αV+α2=0.\Lambda_{\alpha}: \quad U+\alpha V+\alpha^2=0.

The lines corresponding to α\alpha and α′\alpha' meet at (αα′,−α−α′)(\alpha\alpha',-\alpha-\alpha'). No three are concurrent, since a monic quadratic cannot have three distinct roots.

Lemma 2.2. Evaluation at (u,v)(u,v) gives an isomorphism of Q\mathbb{Q}-vector spaces

Q[U,V]≤6→∼L\mathbb{Q}[U,V]_{\leq6}\xrightarrow{\sim}L

and with:

ℓ=(1,u,v,u2,uv,v2)∈L6,(7)\ell=(1,u,v,u^2,uv,v^2)\in L^6, \tag*{(7)}

the evaluation map Hd→LH_d\to L, E↦E(ℓ)E\mapsto E(\ell), is surjective for every d≥3d\geq3.

Proof. Suppose that g∈Q[U,V]≤6g\in\mathbb{Q}[U,V]_{\leq6} vanishes at (u,v)(u,v). Lemma 2.1 implies that it vanishes at all 28 intersections. Each line Λα\Lambda_\alpha contains seven distinct intersections, so the restriction of gg to that line is zero. Over the splitting field, the product of the eight distinct line equations therefore divides gg. Since deg⁡g≤6\deg g\leq6, we obtain g=0g=0. Both vector spaces have dimension 28, proving the first assertion.

Every monomial UiVjU^iV^j with i+j≤6i+j\leq6 is a product of at most three elements of {U,V,U2,UV,V2}\{U,V,U^2,UV,V^2\}. Padding with powers of x0x_0 gives a homogeneous cubic restricting to this monomial under

(x0,…,x5)=(1,U,V,U2,UV,V2).(x_0,\ldots,x_5)=(1,U,V,U^2,UV,V^2).

Multiplication by x0d−3x_0^{d-3} proves the assertion for all d≥3d\geq3. □

We use the ordered basis

B=(UiVa−i:0≤a≤6, 0≤i≤a),(8)\mathcal{B}=(U^iV^{a-i}:0\leq a\leq6,\ 0\leq i\leq a), \tag*{(8)}

where aa and then ii increase, and where the entries are evaluated at (u,v)(u,v) when regarded as elements of LL. For exact computations, one can instead use the relations

r0=U4−6U3V2+5U2V4−UV6−1,r_0=U^4-6U^3V^2+5U^2V^4-UV^6-1,
r1=4U3V−10U2V3+6UV5−V7−1.(9)r_1=4U^3V-10U^2V^3+6UV^5-V^7-1. \tag*{(9)}

Indeed, division of h(z)h(z) by z2+Vz+Uz^2+Vz+U leaves remainder r0+zr1r_0+zr_1. The quotient Q[U,V]/(r0,r1)\mathbb{Q}[U,V]/(r_0,r_1) then identifies with LL; the verification is given in Appendix § 9, by giving a Gröbner basis and proving that this quotient has dimension 28, and thus that its natural surjection to LL is an isomorphism.

The explicit formal curve

Retain the field L=Q(u,v)L = \mathbb{Q}(u,v) from Lemma 2.1 and its ordered basis of monomials of degree at most six. Define the resultant:

H(U,V)=Res⁡z(z8−z−1,z2+Vz+U).(10)H(U,V)=\operatorname{Res}_{z}\left(z^8-z-1,z^2+Vz+U\right). \tag*{(10)}

Explicitly:

H(U,V)=1+U−V−2U4+16U3V2−7U4V+14U3V3−20U2V4+8UV6−7U2V5+U8+UV7−V8.\begin{aligned} H(U,V) ={}&1+U-V-2U^4+16U^3V^2-7U^4V+14U^3V^3\\ &-20U^2V^4+8UV^6-7U^2V^5+U^8+UV^7-V^8. \end{aligned}

If α\alpha ranges over the eight roots of z8−z−1z^8-z-1 in a splitting field, then

H(U,V)=∏α(U+αV+α2).H(U,V)=\prod_{\alpha}(U+\alpha V+\alpha^2).

We use the affine coordinates

(x0,x1,x2,x3,x4,x5)=(1,U,V,U2+A,UV+B,V2+C).(11)(x_0,x_1,x_2,x_3,x_4,x_5)=(1,U,V,U^2+A,UV+B,V^2+C). \tag*{(11)}

For a homogeneous polynomial GG in the variables x0,…,x5x_0,\ldots,x_5, write GchG_{\mathrm{ch}} for its substitution under (11). Direct expansion gives

Dch=AC−B2,Wch(U,V,0,0,0)=H(U,V).(12)D_{\mathrm{ch}}=AC-B^2,\qquad W_{\mathrm{ch}}(U,V,0,0,0)=H(U,V). \tag*{(12)}

The three normal derivatives a,b,ca,b,c of WchW_{\mathrm{ch}} on A=B=C=0A=B=C=0, namely a=∂AWch∣A=B=C=0a=\partial_AW_{\mathrm{ch}}|_{A=B=C=0}, and similarly for bb and cc, are the following polynomials:

a(U,V)=4U6−14U2V−4U2+14UV3+16UV2−7V5−20V4,a(U,V)=4U^6-14U^2V-4U^2+14UV^3+16UV^2-7V^5-20V^4,
b(U,V)=14U2V2+V6,b(U,V)=14U^2V^2+V^6,
c(U,V)=14U3V+16U3−14U2V3−40U2V2+3UV5+24UV4−4V6.(13)c(U,V)=14U^3V+16U^3-14U^2V^3-40U^2V^2+3UV^5+24UV^4-4V^6. \tag*{(13)}

Set

K=ac−b24,K=ac-\frac{b^2}{4},
V0=−231U6+11692U2V+215U2−8332UV3−988UV2+14V5+411V4.(14)V_0=-231U^6+\frac{1169}{2}U^2V+215U^2-\frac{833}{2}UV^3-988UV^2+\frac{1}{4}V^5+411V^4. \tag*{(14)}

Lemma 3.1. At (u,v)(u,v) one has

H=HU=HV=0,V0−K=0,H=H_U=H_V=0,\qquad V_0-K=0,

and Hess⁡(H)(u,v)\operatorname{Hess}(H)(u,v) is invertible.

Proof. The point (u,v)=(βγ,−β−γ)(u,v)=(\beta\gamma,-\beta-\gamma) is the intersection of the two lines indexed by β\beta and γ\gamma. No third line passes through it. Let c∗c_* be the product of the other six line equations, evaluated at this point. Then c∗≠0c_*\ne0, and differentiating the product of the eight line equations gives

det⁡Hess⁡(H)(u,v)=−c∗2(β−γ)2≠0.\det\operatorname{Hess}(H)(u,v)=-c_*^2(\beta-\gamma)^2\ne0.

The same product description gives the three stated vanishings of HH. Finally, expansion of K=ac−b2/4K = ac - b^2/4 using (13) and then reduction by the two relations (9) gives

ac−b24≡V0(mod(r0,r1)).ac - \frac{b^2}{4} \equiv V_0 \pmod{(r_0,r_1)}.

The displayed reduction basis in the finite-certificate appendix makes this a rational polynomial identity; the supplementary verifier checks it independently. This shows that V0−KV_0-K vanishes at (u,v)(u,v). □

All derivatives in the next definition are taken as polynomial derivatives in U,VU,V before evaluation at (u,v)(u,v). Define u1,v1∈Lu_1,v_1 \in L by

\begin{pmatrix} u_1 \\ v_1 \end{pmatrix} = -\operatorname{Hess}(H)(u,v)^{-1}\nabla(V_0-K)(u,v). \tag*{(15)}\end{pmatrix}

Put

ℓ=(1,u,v,u2,uv,v2),w0=0,w1=u1,w2=v1,\ell=(1,u,v,u^2,uv,v^2), \qquad w_0=0,\qquad w_1=u_1,\qquad w_2=v_1,
w3=2uu1−c(u,v),w4=vu1+uv1+12b(u,v),w_3=2uu_1-c(u,v),\qquad w_4=vu_1+uv_1+\frac{1}{2}b(u,v),
w5=2vv1−a(u,v).(16)w_5=2vv_1-a(u,v). \tag*{(16)}

These formulas specify a continuous Q\mathbb{Q}-algebra homomorphism from RR to L[[t]]L[[t]], i.e. an arc:

φ:R=Q[[x0,…,x5]]⟶L[[t]],xi↦tℓi+t2wi,P=ker⁡φ.(17)\varphi:R=\mathbb{Q}[[x_0,\ldots,x_5]]\longrightarrow L[[t]],\qquad x_i\mapsto t\ell_i+t^2w_i,\qquad P=\ker\varphi. \tag*{(17)}

(Here ℓi\ell_i is the ii-th coordinate of ℓ\ell.) In particular, φ(x0)=t\varphi(x_0)=t. Geometrically φ\varphi corresponds to a formal curve in the formal spectrum of RR.

Lemma 3.2. For every d≥3d\geq3,

φ(md)=tdL[[t]].(18)\varphi(\mathfrak{m}^d)=t^dL[[t]]. \tag*{(18)}

The image A=φ(R)A=\varphi(R) is a one-dimensional complete local domain, finite over Q[[t]]\mathbb{Q}[[t]], with residue field Q\mathbb{Q}. Its fraction field is L((t))L((t)), and its normalization is L[[t]]L[[t]]. Consequently, PP is a height-five prime ideal and P⊆m2P\subseteq\mathfrak{m}^2.

Proof. By Lemma 2.2, choose homogeneous cubics b1,…,b28b_1,\ldots,b_{28} such that bj(ℓ)b_j(\ell) form a Q\mathbb{Q}-basis of LL. Relative to this basis, the elements t−3φ(bj)t^{-3}\varphi(b_j) form the columns of a matrix in Mat⁡28(Q[[t]])\operatorname{Mat}_{28}(\mathbb{Q}[[t]]) whose reduction modulo tt is invertible. They therefore form a Q[[t]]\mathbb{Q}[[t]]-basis of L[[t]]L[[t]]. Since φ(x0)=t\varphi(x_0)=t, every element of t3L[[t]]t^3L[[t]] lifts to an element of m3\mathfrak{m}^3. Multiplication by x0d−3x_0^{d-3} proves the reverse inclusion in (18); the forward inclusion follows from φ(xi)∈tL[[t]]\varphi(x_i)\in tL[[t]].

The image AA contains Q[[t]]\mathbb{Q}[[t]] and is a Q[[t]]\mathbb{Q}[[t]]-submodule of the finite free module L[[t]]L[[t]]. It is therefore finite over Q[[t]]\mathbb{Q}[[t]]. It is a domain of dimension one. As a quotient of the complete local ring RR, it is complete and local. Reduction modulo tt takes AA onto Q\mathbb{Q}; its kernel is the image of m\mathfrak{m}, so its residue field is Q\mathbb{Q}. The inclusion t3L[[t]]⊆At^3L[[t]]\subseteq A implies that Frac⁡(A)=L((t))\operatorname{Frac}(A)=L((t)). The ring L[[t]]L[[t]] is finite over AA and integrally closed in this field, hence is the normalization of AA.

It follows that ht⁡P=6−1=5\operatorname{ht} P = 6 - 1 = 5. The six coordinates of ℓ\ell are linearly independent over Q\mathbb{Q} by Lemma 2.2. The coefficient of tt in φ(p)\varphi(p) therefore excludes a nonzero linear initial form for any p∈Pp \in P. The constant coefficient also vanishes, so P⊆m2P \subseteq\mathfrak{m}^{2}.

Eliminating critical values to all orders

Consider the homogeneous cubic

V0♯=−231x33+11692x0x2x3+215x02x3−8332x0x4x5−988x0x1x5+14x2x52+411x0x52.(19)V_0^{\sharp} = -231x_3^3 + \frac{1169}{2}x_0x_2x_3 + 215x_0^2x_3 - \frac{833}{2}x_0x_4x_5 - 988x_0x_1x_5 + \frac{1}{4}x_2x_5^2 + 411x_0x_5^2. \tag*{(19)}

Its restriction to A=B=C=0A = B = C = 0 under (11) is exactly V0(U,V)V_0(U,V). Set E5=x02V0♯E_5 = x_0^2V_0^{\sharp}.

Proposition 4.1. There exist homogeneous forms En∈HnE_n \in H_n for all n≥6n \ge6 and a continuous Q\mathbb{Q}-algebra homomorphism ψ:R→L[[t]]\psi: R \to L[[t]], i.e. an arc, such that with:

F:=D+W+E5+∑n≥6En,(20)F := D + W + E_5 + \sum_{n \ge6} E_n, \tag*{(20)}

one has:

ψ(x0)=t,ψ(xi)=tℓi+t2wi+O(t3)(1≤i≤5),(21)\psi(x_0) = t,\qquad\psi(x_i) = t\ell_i + t^2w_i + O(t^3)\qquad(1 \le i \le5), \tag*{(21)}

the critical condition:

ψ(∂iF)=0(0≤i≤5),(22)\psi(\partial_iF) = 0\qquad(0 \le i \le5), \tag*{(22)}

and

ψ(F)=0.\psi(F) = 0.

Proof. We first associate to every (tentative) F=D+W+E5+O(m6)F = D + W + E_5 + O(\mathfrak{m}^6) an arc ψF:R→L[[t]]\psi_F : R \to L[[t]] that satisfies the condition:

ψF(∂iF)=0(1≤i≤5).\psi_F(\partial_iF) = 0\qquad(1 \le i \le5).

First, make the following substitution (the affine cone version of (11)):

x0=t,(x1,…,x5)=t(U,V,U2+A,UV+B,V2+C)x_0 = t,\qquad(x_1,\ldots,x_5) = t(U,V,U^2 + A,UV + B,V^2 + C)

into FF, and then divide by t3t^3; the resulting formal series GFG_F is then:

GF=AC−B2+tWch+t2(E5)ch+O(t3).(23)G_F = AC - B^2 + tW_{\mathrm{ch}} + t^2(E_5)_{\mathrm{ch}} + O(t^3). \tag*{(23)}

We work over L[[t,U−u,V−v,A,B,C]]L[[t,U-u,V-v,A,B,C]] (and thus GF∈L[[t,U−u,V−v,A,B,C]]G_F \in L[[t,U-u,V-v,A,B,C]]). The Hessian of AC−B2AC - B^2 in the three variables A,B,CA,B,C is invertible. The formal implicit function theorem then gives unique critical solutions in tL[[t,U−u,V−v]]tL[[t,U-u,V-v]] satisfying:

AF=−tc+O(t2),BF=12tb+O(t2),CF=−ta+O(t2).(24)A_F=-tc+O(t^2),\qquad B_F=\frac{1}{2}tb+O(t^2),\qquad C_F=-ta+O(t^2). \tag*{(24)}

Here a,b,ca,b,c are the polynomials in U,VU,V as before; the corresponding critical equations are:

∂AGF(t,U,V,AF,BF,CF)=0,∂BGF(t,U,V,AF,BF,CF)=0,∂CGF(t,U,V,AF,BF,CF)=0.\begin{aligned} \partial_{A}G_F(t,U,V,A_F,B_F,C_F)&=0,\\ \partial_{B}G_F(t,U,V,A_F,B_F,C_F)&=0,\\ \partial_{C}G_F(t,U,V,A_F,B_F,C_F)&=0. \end{aligned}

Let G~F∈tL[[t,U−u,V−v]]\widetilde{G}_F\in tL[[t,U-u,V-v]] be the result of substituting these three solutions into GFG_F. A calculation through degree two in tt gives

G~F(t,U,V)=tH(U,V)+t2(V0−K)(U,V)+O(t3).(25)\widetilde{G}_F(t,U,V)=tH(U,V)+t^2(V_0-K)(U,V)+O(t^3). \tag*{(25)}

Put ΨF:=t−1G~F∈L[[t,U−u,V−v]]\Psi_F:=t^{-1}\widetilde{G}_F\in L[[t,U-u,V-v]]. Lemma 3.1 and the formal implicit function theorem give unique series UF,VF∈L[[t]]U_F,V_F\in L[[t]] with constant terms u,vu,v satisfying the critical equations:

∂UΨF(t,UF,VF)=∂VΨF(t,UF,VF)=0.\partial_U\Psi_F(t,U_F,V_F)=\partial_V\Psi_F(t,U_F,V_F)=0.

Their first terms are

UF=u+tu1+O(t2),VF=v+tv1+O(t2),(26)U_F=u+tu_1+O(t^2),\qquad V_F=v+tv_1+O(t^2), \tag*{(26)}

by (15).

For abbreviations we write:

A~F=AF(t,UF,VF),B~F=BF(t,UF,VF),C~F=CF(t,UF,VF),\begin{aligned} \widetilde{A}_F&=A_F(t,U_F,V_F),\\ \widetilde{B}_F&=B_F(t,U_F,V_F),\\ \widetilde{C}_F&=C_F(t,U_F,V_F), \end{aligned}

for the elements of L[[t]]L[[t]] obtained by substituting UF,VFU_F,V_F into the variables U,VU,V.

Substituting UF,VF,A~F,B~F,C~F∈L[[t]]U_F,V_F,\widetilde{A}_F,\widetilde{B}_F,\widetilde{C}_F\in L[[t]] into the variables U,V,A,B,CU,V,A,B,C, we then have the formulas

x0=t,(x1,…,x5)=t(UF,VF,UF2+A~F,UFVF+B~F,VF2+C~F)x_0=t,\qquad(x_1,\ldots,x_5)=t(U_F,V_F,U_F^2+\widetilde{A}_F,U_FV_F+\widetilde{B}_F,V_F^2+\widetilde{C}_F)

which then defines an arc ψF:R→L[[t]]\psi_F:R\to L[[t]].

Now (in the affine chart) the change of coordinate map

(U,V,A,B,C)⟼(U,V,U2+A,UV+B,V2+C)(U,V,A,B,C)\longmapsto(U,V,U^2+A,UV+B,V^2+C)

has determinant-one Jacobian. The five critical equations that the five critical solutions AF,BF,CF,UF,VFA_F,B_F,C_F,U_F,V_F satisfy, thus give:

ψF(∂iF)=0(1≤i≤5).(27)\psi_F(\partial_iF)=0\qquad(1\leq i\leq5). \tag*{(27)}

Note also that equations (24) and (26) give precisely the coefficients wiw_i in (16). Thus every such ψF\psi_F satisfies (21).

Now define the critical value V(F)\mathcal{V}(F) of ψF\psi_F by:

V(F)=ψF(F)∈L[[t]].\mathcal{V}(F)=\psi_F(F)\in L[[t]].

We thus have by definition:

V(F)=t3G~F(t,UF,VF).\mathcal{V}(F)=t^3\widetilde{G}_F(t,U_F,V_F).

The vanishings in Lemma 3.1 give:

H(UF,VF)=O(t2),(V0−K)(UF,VF)=O(t)H(U_F,V_F)=O(t^2),\qquad(V_0-K)(U_F,V_F)=O(t)

and hence (25) then implies

V(F)∈t6L[[t]].(28)\mathcal{V}(F)\in t^6L[[t]]. \tag*{(28)}

Next we claim, for a homogeneous E∈HnE\in H_n, n≥6n\ge6, that

V(F+E)−V(F)≡tnE(ℓ)(modtn+1).(29)\mathcal{V}(F+E)-\mathcal{V}(F)\equiv t^nE(\ell)\pmod{t^{n+1}}. \tag*{(29)}

For this, introduce a parameter ss and carry out the two recursive implicit constructions for F+sEF+sE. Each coefficient of the resulting arc lies in L[s]L[s]: the constant matrices to be inverted are unchanged, and all subsequent recursion uses polynomial operations and these fixed inverse matrices. Thus the chain rule applies coefficientwise in L[s][[t]]L[s][[t]]. Since x0=tx_0=t is fixed and the five spatial derivatives vanish, it gives

ddsV(F+sE)=ψF+sE(E)≡tnE(ℓ)(modtn+1).\frac{d}{ds}\mathcal{V}(F+sE)=\psi_{F+sE}(E)\equiv t^nE(\ell)\pmod{t^{n+1}}.

Integrating from s=0s=0 to s=1s=1 proves (29). Note that this is also a crucial point that makes use of the fact that the coefficient field has characteristic zero.

Now start with F[5]=D+W+E5F^{[5]}=D+W+E_5. Inductively suppose that F[n−1]F^{[n-1]} has been chosen and that its critical value has no term of degree less than nn. Let cn∈Lc_n\in L be the coefficient of tnt^n. By Lemma 2.2, choose En∈HnE_n\in H_n with En(ℓ)=−cnE_n(\ell)=-c_n. Equation (29) shows that F[n]=F[n−1]+EnF^{[n]}=F^{[n-1]}+E_n kills the coefficient of tnt^n while preserving all previous vanishings.

Now adding a form of degree nn changes GFG_F and its normal equations by O(tn−3)O(t^{n-3}). The fixed invertible normal Hessian then shows that the normal critical solutions, as functions of t,U,Vt,U,V, change by O(tn−3)O(t^{n-3}). After reduction and division by tt, the tangential equations change by O(tn−4)O(t^{n-4}). Their fixed invertible constant Hessian shows that UF,VFU_F,V_F change by O(tn−4)O(t^{n-4}). Since the normal solutions and their U,VU,V derivatives are divisible by tt, their evaluations on the final tangential arc change by O(tn−3)O(t^{n-3}). The original coordinates x1,…,x5x_1,\ldots,x_5 therefore change by O(tn−3)O(t^{n-3}). These orders tend to infinity with nn.

It follows that F[n]F^{[n]} converges mm-adically to the series FF with the shape in (20), and the arcs converge coefficientwise to an arc ψ=ψF\psi=\psi_F satisfying (21). Substitution and formal differentiation are continuous for these filtrations, so the induction and (27) give

ψ(F)=0,ψ(∂iF)=0(1≤i≤5).\psi(F)=0,\qquad\psi(\partial_iF)=0\quad(1\le i\le5).

Finally to complete the proof of (22), it remains to show that ψ(∂0F)=0\psi(\partial_0F)=0. Differentiate ψ(F)=0\psi(F)=0 with respect to tt. The chain rule and ψ(x0)=t\psi(x_0)=t give

0=ψ(∂0F)+∑i=15ψ(∂iF)dψ(xi)dt=ψ(∂0F),0=\psi(\partial_0F)+\sum_{i=1}^{5}\psi(\partial_iF)\frac{d\psi(x_i)}{dt}=\psi(\partial_0F),

which thus completes the proof of (22). □\square

Straightening and symbolic-square membership

Lemma 5.1. Let ψ:R→L[[t]]\psi:R\to L[[t]] be any arc that satisfies (21). There is a continuous Q\mathbb{Q}-algebra automorphism θ\theta of RR such that

θ(x0)=x0,θ(xi)−xi∈m3(1≤i≤5),φ∘θ=ψ.\theta(x_0)=x_0,\qquad\theta(x_i)-x_i\in\mathfrak{m}^{3}\quad(1\le i\le5),\qquad\varphi\circ\theta=\psi.

Proof. Each difference ψ(xi)−φ(xi)\psi(x_i)-\varphi(x_i) belongs to t3L[[t]]t^3L[[t]]. By (18), choose gi∈m3g_i\in\mathfrak{m}^3 whose image under φ\varphi is this difference. Define θ(xi)=xi+gi\theta(x_i)=x_i+g_i for 1≤i≤51\le i\le5 and θ(x0)=x0\theta(x_0)=x_0. This substitution has identity linear part, so recursive solution of its homogeneous terms gives a continuous inverse. Its composite with φ\varphi agrees with ψ\psi on the six variables and hence on RR by continuity. □\square

Apply this lemma to the arc of Proposition 4.1, and put

f=θ(F).(30)f=\theta(F). \tag*{(30)}

Changing variables by terms of order at least three changes a cubic only in degrees at least five. Therefore

f≡D+W(modm5).(31)f\equiv D+W\pmod{\mathfrak{m}^{5}}. \tag*{(31)}

The chain rule and φ∘θ=ψ\varphi\circ\theta=\psi imply

φ(f)=0,φ(∂if)=0(0≤i≤5).(32)\varphi(f)=0,\qquad\varphi(\partial_i f)=0\quad(0\le i\le5). \tag*{(32)}

Proposition 5.2. The series ff in (30) belongs to P(2)P^{(2)}. More generally, if g∈Rg\in R satisfies φ(g)=φ(∂ig)=0\varphi(g)=\varphi(\partial_i g)=0 for 1≤i≤51\le i\le5, then g∈P(2)g\in P^{(2)}.

Proof. Let S=L[[x0,…,x5]]S=L[[x_0,\ldots,x_5]] and set

Qarc=(xi−x0ℓi−x02wi:1≤i≤5)S.Q_{\mathrm{arc}}=(x_i-x_0\ell_i-x_0^2w_i:1\le i\le5)S.

The same formulas that define φ:R→L[[t]]\varphi:R\to L[[t]] as before then give a surjective homomorphism S→L[[t]]S\to L[[t]] with kernel QarcQ_{\mathrm{arc}}, and Qarc∩R=PQ_{\mathrm{arc}}\cap R=P. Use x0x_0 and yi:=xi−x0ℓi−x02wiy_i:=x_i-x_0\ell_i-x_0^2w_i (1≤i≤5)(1\le i\le5) as formal coordinates on SS. In terms of these new coordinates, we have S=L[[x0]][[y1,…,y5]]S=L[[x_0]][[y_1,\ldots,y_5]], and then QarcQ_{\mathrm{arc}} is the ideal of SS generated by yiy_i for 1≤i≤51\le i\le5, and hence Qarc2Q_{\mathrm{arc}}^2 is the ideal of SS generated by yiyjy_i y_j for 1≤i,j≤51\le i,j\le5. One then sees immediately, by using Taylor expansion for example (with respect to the original coordinates x1,…,x5x_1,\ldots,x_5, with x0x_0 being regarded as fixed), that if g∈Rg\in R satisfies φ(g)=φ(∂ig)=0\varphi(g)=\varphi(\partial_i g)=0 for 1≤i≤51\le i\le5, then g∈Qarc2g\in Q_{\mathrm{arc}}^2. Since L/QL/\mathbb{Q} is finite and separable, S=R⊗QLS=R\otimes_{\mathbb{Q}}L is finite étale over RR. The local map RP→SQarcR_P\to S_{Q_{\mathrm{arc}}} is faithfully flat. Its closed fiber is a localization of a finite étale algebra over κ(P)\kappa(P), hence a field. Consequently,

PSQarc=QarcSQarc.PS_{Q_{\mathrm{arc}}}=Q_{\mathrm{arc}}S_{Q_{\mathrm{arc}}}.

Faithful flatness therefore gives

g∈P2SQarc∩RP=P2RP.g\in P^2S_{Q_{\mathrm{arc}}}\cap R_P=P^2R_P.

Intersecting with RR proves g∈P(2)g\in P^{(2)}. Finally the particular series ff satisfies the hypotheses by (32). □

A finite certificate for nonmembership

Put

T=Q[[x1,…,x5]],q=(x1,…,x5)T,J=(P+(x0))/(x0)⊂T.T=\mathbb{Q}[[x_1,\ldots,x_5]],\qquad q=(x_1,\ldots,x_5)T,\qquad J=(P+(x_0))/(x_0)\subset T.

A bar denotes the image under the substitution x0=0x_0=0 (here we identify R/(x0)R/(x_0) with TT). We shall construct a Q\mathbb{Q}-linear functional λ:T⟶Q\lambda:T\longrightarrow\mathbb{Q} such that

λ(qJ)=0,λ(q5‾)=0,λ(D‾+W‾)=1.(33)\lambda(qJ)=0,\qquad\lambda(\overline{q^5})=0,\qquad\lambda(\overline{D}+\overline{W})=1. \tag*{(33)}

Together with (31), these identities prove the required nonmembership in mPmP.

Consider the six quadrics:

q1=x0x3−x12,q2=x0x4−x1x2,q3=x0x5−x22,q4=x1x4−x2x3,q5=x1x5−x2x4,q6=x3x5−x42.(34)\begin{aligned} q_1&=x_0x_3-x_1^2,\qquad q_2=x_0x_4-x_1x_2,\qquad q_3=x_0x_5-x_2^2,\\ q_4&=x_1x_4-x_2x_3,\qquad q_5=x_1x_5-x_2x_4,\qquad q_6=x_3x_5-x_4^2. \tag*{(34)} \end{aligned}

with their accompanying homogeneous cubic terms given by:

C1=16x0x1x3+14x0x3x4−40x0x3x5+24x1x52−14x2x3x5+3x4x52−4x53,C2=−7x0x3x5−12x53,C3=−4x02x3+16x0x1x5−14x0x2x3+14x0x4x5−20x0x52−7x2x52+4x33,C4=−12x03−72x0x32−72x3x52,C5=−12x03−8x0x3x5−4x53,C6=4x03+3x02x1−3x02x2−4x0x32+16x1x3x5−7x2x32+14x3x4x5−20x3x52.(35)\begin{aligned} C_1={}&16x_0x_1x_3+14x_0x_3x_4-40x_0x_3x_5+24x_1x_5^2\\ &-14x_2x_3x_5+3x_4x_5^2-4x_5^3,\\ C_2={}&-7x_0x_3x_5-\frac{1}{2}x_5^3,\\ C_3={}&-4x_0^2x_3+16x_0x_1x_5-14x_0x_2x_3+14x_0x_4x_5\\ &-20x_0x_5^2-7x_2x_5^2+4x_3^3,\\ C_4={}&-\frac{1}{2}x_0^3-\frac{7}{2}x_0x_3^2-\frac{7}{2}x_3x_5^2,\\ C_5={}&-\frac{1}{2}x_0^3-8x_0x_3x_5-4x_5^3,\\ C_6={}&4x_0^3+3x_0^2x_1-3x_0^2x_2-4x_0x_3^2+16x_1x_3x_5\\ &-7x_2x_3^2+14x_3x_4x_5-20x_3x_5^2. \tag*{(35)} \end{aligned}

Lemma 6.1. The image of PP in R/m4R/\mathfrak{m}^4 is the rational span of the classes of

qi+Ci(1≤i≤6),xjqi(0≤j≤5, 1≤i≤6).(36)q_i+C_i\quad(1\le i\le6),\qquad x_jq_i\quad(0\le j\le5,\ 1\le i\le6). \tag*{(36)}

Proof. Let ρ\rho denote substitution of (1,U,V,U2,UV,V2)(1,U,V,U^2,UV,V^2) into homogeneous forms. The degree-two kernel of ρ\rho is spanned by the six qiq_i, and its degree-three kernel is spanned by their linear multiples. These statements can be seen by homogenizing the substitution to the map

(x0,x1,x2,x3,x4,x5)⟼(s2,sr,ss′,r2,rs′,s′2).(x_0,x_1,x_2,x_3,x_4,x_5)\longmapsto(s^2,sr,ss',r^2,rs',s'^2).

The six quadrics in (34) generate all relations among these degree-two monomials: any two products representing the same monomial in s,r,s′s,r,s' are related by changing the pairing of four factors. The quotient dimensions in degrees two and three are therefore 15 and 28, respectively. Since ρ(Hd)\rho(H_d) consists of polynomials of degree at most 2d2d, Lemma 2.2 shows that evaluation at ℓ\ell has these same kernels for d=2,3d=2,3.

By the definition of φ\varphi as given in (17), the coefficient of t3t^3 in φ(qi)\varphi(q_i) is

∑j=05∂jqi(ℓ)wj.\sum_{j=0}^{5}\partial_jq_i(\ell)w_j.

The negatives of these six coefficients, in order, are

c,−12b,a,−(12ub+vc),ua+12vb,u2a+uvb+v2c,(37)c,\quad-\tfrac{1}{2}b,\quad a,\quad-\left(\tfrac{1}{2}ub+vc\right),\quad ua+\tfrac{1}{2}vb,\quad u^2a+uvb+v^2c, \tag*{(37)}

where a,b,ca,b,c are evaluated at (u,v)(u,v). The terms containing u1,v1u_1,v_1 cancel. Substitution into the displayed CiC_i, followed by reduction using (9), gives exactly the values in (37). Consequently,

φ(qi+Ci)∈t4L[[t]](1≤i≤6).(38)\varphi(q_i+C_i)\in t^4L[[t]]\qquad(1\leq i\leq6). \tag*{(38)}

For p∈Pp\in P, Lemma 3.2 gives p=p2+p3+O(m4)p=p_2+p_3+O(\mathfrak m^4) with pd∈Hdp_d\in H_d. The coefficient of t2t^2 in φ(p)\varphi(p) gives p2=∑iαiqip_2=\sum_i\alpha_iq_i for some αi∈Q\alpha_i\in\mathbb{Q}. By (38), the cubic p3−∑iαiCip_3-\sum_i\alpha_iC_i evaluates to zero at ℓ\ell. It is therefore a linear combination of the xjqix_jq_i, proving one inclusion.

Conversely, every expression in (36) maps into t4L[[t]]t^4L[[t]]. Lemma 3.2 gives φ(m4)=t4L[[t]]\varphi(\mathfrak m^4)=t^4L[[t]], so its image can be canceled by an element of m4\mathfrak m^4. Its class modulo m4\mathfrak m^4 thus belongs to the image of PP.

Now define the functional λ\lambda on monomials by Table 1, and set it to zero on every monomial not listed. It extends to all of TT by taking this finite linear combination of coefficients; in particular, λ(q5)=0\lambda(q^5)=0.

Degree-three tupleValueDegree-four tupleValue
(0,0,0,2,1)(0,0,0,2,1)−5632/49-5632/49(0,0,0,0,4)(0,0,0,0,4)33
(0,0,0,3,0)(0,0,0,3,0)−10240/343-10240/343(0,0,0,1,3)(0,0,0,1,3)88
(0,1,0,1,1)(0,1,0,1,1)−12-12(0,0,0,2,2)(0,0,0,2,2)384/49384/49
(0,1,0,2,0)(0,1,0,2,0)−32-32(0,0,0,3,1)(0,0,0,3,1)512/343512/343
(0,1,1,0,1)(0,1,1,0,1)−192/7-192/7(0,0,1,0,3)(0,0,1,0,3)384/49384/49
(0,1,1,1,0)(0,1,1,1,0)−1536/49-1536/49(0,0,1,1,2)(0,0,1,1,2)512/343512/343
(0,2,0,1,0)(0,2,0,1,0)−3/2-3/2(0,0,3,1,0)(0,0,3,1,0)−3/8-3/8
(0,2,1,0,0)(0,2,1,0,0)−8-8(0,1,0,2,1)(0,1,0,2,1)8/78/7
(1,0,0,2,0)(1,0,0,2,0)−1280/49-1280/49(0,1,0,3,0)(0,1,0,3,0)64/4964/49
(1,1,0,0,1)(1,1,0,0,1)−3/2-3/2(0,1,1,0,2)(0,1,1,0,2)8/78/7
(1,1,0,1,0)(1,1,0,1,0)−4-4(0,1,1,1,1)(0,1,1,1,1)64/4964/49
(1,1,1,0,0)(1,1,1,0,0)−192/49-192/49(1,0,0,1,2)(1,0,0,1,2)8/78/7
(2,0,0,0,1)(2,0,0,0,1)−4-4(1,0,0,2,1)(1,0,0,2,1)64/4964/49
(2,0,0,1,0)(2,0,0,1,0)32/4932/49(1,0,1,0,2)(1,0,1,0,2)64/4964/49
(2,0,1,0,0)(2,0,1,0,0)1536/3431536/343
(3,0,0,0,0)(3,0,0,0,0)24/724/7

Table 1. All nonzero values of λ\lambda. An exponent tuple e=(e1,…,e5)e=(e_1,\ldots,e_5) denotes x1e1⋯x5e5x_1^{e_1}\cdots x_5^{e_5}.

Proposition 6.2. The functional λ\lambda satisfies (33). Every series f≡D+W(modm5)f\equiv D+W\pmod{\mathfrak m^5} consequently lies outside mP\mathfrak mP.

Proof. The displayed coefficients give the rational identities

λ(xj(qi+Ci))=0(1≤j≤5, 1≤i≤6),(39)\lambda\bigl(x_j(q_i+C_i)\bigr)=0\qquad(1\leq j\leq5,\ 1\leq i\leq6), \tag*{(39)}
λ(xjxkqi)=0(1≤j≤5, 0≤k≤5, 1≤i≤6).(40)\lambda(x_jx_kq_i)=0\qquad(1\leq j\leq5,\ 0\leq k\leq5,\ 1\leq i\leq6). \tag*{(40)}

These are 30+18030+180 finite identities; all are checked by the exact-arithmetic verifier included with the supplementary material. Note that of course for (40) it is trivial when k=0k=0, so that the number of non-trivial identities in (40) is 150, and the total number of non-trivial identities that have to be checked in (39)–(40) is 180.

Note also how the dimension count works here: the dimension of the space of cubics in five variables is 56, and the dimension of the space of quartics in five variables is 126; as (56+126)−180=2(56+126)-180=2, so this leaves two degrees of freedom to define the functional λ\lambda satisfying (39)–(40) (these two remaining degrees of freedom are used to make λ(D‾)=4\lambda(\overline{D})=4 and λ(W‾)=−3\lambda(\overline{W})=-3 as shown below; incidentally this also explains why λ\lambda is defined this way).

For g∈qJg\in\mathfrak qJ, choose pj∈Pp_j\in P such that g=∑j=15xjpjg=\sum_{j=1}^{5}x_jp_j. The terms of gg through degree four depend only on the classes of the pjp_j modulo m4\mathfrak m^4. Lemma 6.1 and (39)–(40) therefore give λ(g)=0\lambda(g)=0. This proves vanishing of λ\lambda on the whole ideal qJ\mathfrak qJ.

Finally, (2) and (3) give

D‾=2x1x2x4−x12x5−x22x3\overline{D}=2x_1x_2x_4-x_1^2x_5-x_2^2x_3

and

W‾=8x1x53−7x2x3x52+x34+x4x53−x54.\overline{W}=8x_1x_5^3-7x_2x_3x_5^2+x_3^4+x_4x_5^3-x_5^4.

Table 1 gives λ(D‾)=4\lambda(\overline{D})=4 and λ(W‾)=−3\lambda(\overline{W})=-3. Thus λ(f)=1\lambda(f)=1 whenever f≡D+W(modm5)f\equiv D+W\pmod{\mathfrak m^5}. If f∈mPf\in\mathfrak mP, then f∈qJf\in\mathfrak qJ, a contradiction. □\square

Proof of Theorem 1.1. The field LL and φ:R→L[[t]]\varphi:R\to L[[t]] are specified in (5) and (15)–(17). Lemma 3.2 proves the assertions about the height, residue field, and normalization. Proposition 4.1 and Lemma 5.1 construct a series ff given in equation (30), with f≡D+W(modm5)f\equiv D+W\pmod{\mathfrak m^5}. Proposition 5.2 gives f∈P(2)f\in P^{(2)}, and Proposition 6.2 gives f∉mPf\notin\mathfrak mP. □\square

Extension of the coefficient field

For an ideal II in a Noetherian ring, recall the symbolic square:

I(2)=⋂q∈Min⁡(I)(I2Rq∩R).I^{(2)}=\bigcap_{\mathfrak{q}\in\operatorname{Min}(I)}(I^2R_{\mathfrak{q}}\cap R).

The prime ideal PP of RR in the previous constructions acquires a particularly simple branch description after extension of the coefficient field.

Proposition 7.1. Let RC=C[[x0,…,x5]]R_C=\mathbb{C}[[x_0,\ldots,x_5]], let mC=(x0,…,x5)RC\mathfrak{m}_C=(x_0,\ldots,x_5)R_C, and put I=PRCI=PR_C. Then II is an unmixed radical ideal, and we have the primary decomposition:

I=⋂σ:L↪C(xi−x0σ(ℓi)−x02σ(wi):1≤i≤5)RC.(41)I=\bigcap_{\sigma:L\hookrightarrow\mathbb{C}}\left(x_i-x_0\sigma(\ell_i)-x_0^2\sigma(w_i):1\leq i\leq5\right)R_C. \tag*{(41)}

These 28 ideals are the distinct smooth branch primes of II, and the series ff of equation (30) satisfies

f∈I(2)∖mCI.f\in I^{(2)}\setminus\mathfrak{m}_CI.

Proof. By Lemma 3.2, the inclusion A⊂L[[t]]A\subset L[[t]] is an inclusion of finite Q[[t]]\mathbb{Q}[[t]]-modules. The maximal-ideal topology on AA agrees with its tt-adic topology: the quotient A/tAA/tA is a finite-dimensional local Q\mathbb{Q}-algebra. Completed extension of the coefficient field consequently gives

RC/PRC≅A⊗Q[[t]]C[[t]].(42)R_C/PR_C\cong A\otimes_{\mathbb{Q}[[t]]}\mathbb{C}[[t]]. \tag*{(42)}

For completeness, this identification can be seen by reducing the finite Q[[t]]\mathbb{Q}[[t]]-module AA modulo tnt^n, extending coefficients to C\mathbb{C}, and taking the inverse limit. Finiteness identifies that limit with the tensor product in (42); the equivalence of topologies identifies it with the completed presentation C[[x]]/PC[[x]]\mathbb{C}[[x]]/P\mathbb{C}[[x]].

The extension Q[[t]]⟶C[[t]]\mathbb{Q}[[t]]\longrightarrow\mathbb{C}[[t]] is flat, since its target is torsion-free over the discrete valuation ring Q[[t]]\mathbb{Q}[[t]]. It follows that

RC/I↪C[[t]]⊗Q[[t]]L[[t]]≅∏σ:L↪CC[[t]].R_C/I\hookrightarrow\mathbb{C}[[t]]\otimes_{\mathbb{Q}[[t]]}L[[t]]\cong\prod_{\sigma:L\hookrightarrow\mathbb{C}}\mathbb{C}[[t]].

The image contains t3t^3 times the product on the right. The coordinate maps have the kernels displayed in (41); their intersection is therefore II. Each quotient is C[[t]]\mathbb{C}[[t]], and distinct embeddings give distinct pairs (σ(u),σ(v))(\sigma(u),\sigma(v)), hence distinct tangent directions. Thus these are precisely the minimal primes of II.

The identities φ(f)=φ(∂if)=0\varphi(f)=\varphi(\partial_i f)=0 persist on every branch. Taylor expansion in the five transverse branch coordinates therefore places ff in the square of each branch ideal, and hence in I(2)I^{(2)}. Extend the functional of Table 1 C\mathbb{C}-linearly on fourth jets. The identities in the proof of Proposition 6.2 remain valid under this extension. The extended functional annihilates the image of mCI\mathfrak{m}_CI modulo x0x_0 and takes value 11 on ff, proving the nonmembership. □

Corollary 7.2. There is a formal power series f∈C[[x0,…,x5]]f \in\mathbb{C}[[x_0,\ldots,x_5]] without constant term such that

f∉mC(f,∂0f,…,∂5f).f \notin\mathfrak{m}_{\mathbb{C}}\sqrt{(f,\partial_0f,\ldots,\partial_5f)}.

Proof. Take the series given by equation (30). The radical ideal II of Proposition 7.1 contains ff and all its partial derivatives, so the displayed radical is contained in II. The assertion follows from f∉mCIf \notin\mathfrak{m}_{\mathbb{C}}I.

This is the formal-power-series problem in [1 Introduction and Corollary 2]. Proposition 7.1 then also gives a counterexample to the general form of the Eisenbud–Mazur conjecture stated in [4 Conjecture 1.5] (with C\mathbb{C} being the coefficient field).

A counterexample in a localized polynomial ring

The formulas that define φ:R→L[[t]]\varphi:R\to L[[t]] also define an algebraic curve. The construction of ff uses a complete formal power series, but by using flatness and a differential characterization of the symbolic square, we are able to construct a counterexample to the original Eisenbud–Mazur conjecture in a localized polynomial ring.

Let

S0=Q[x0,…,x5]n,n=(x0,…,x5),S_0=\mathbb{Q}[x_0,\ldots,x_5]_{\mathfrak n},\qquad\mathfrak n=(x_0,\ldots,x_5),

where the same notation n\mathfrak n denotes the maximal ideal after localization. Let p0\mathfrak p_0 be the kernel of the polynomial homomorphism

Q[x0,…,x5]⟶L[t],xi⟼tℓi+t2wi,\mathbb{Q}[x_0,\ldots,x_5]\longrightarrow L[t],\qquad x_i\longmapsto t^{\ell_i}+t^2w_i,

and put p=p0S0\mathfrak p=\mathfrak p_0S_0.

Theorem 8.1. The ideal p\mathfrak p is a height-five prime of S0S_0. Its extension to the completion R=S0^R=\widehat{S_0} is PP, and

p(2)R=P(2).(43)\mathfrak p^{(2)}R=P^{(2)}. \tag*{(43)}

There exists a polynomial g∈Q[x0,…,x5]g\in\mathbb{Q}[x_0,\ldots,x_5] such that

g≡D+W(mod(x0,…,x5)5),g∈p(2)∖np.(44)g\equiv D+W\pmod{(x_0,\ldots,x_5)^5},\qquad g\in\mathfrak p^{(2)}\setminus\mathfrak n\mathfrak p. \tag*{(44)}

In particular, the counterexample exists in a regular local ring essentially of finite type over Q\mathbb{Q}.

Proof. We first identify the completed prime. Write

B=Q[t,tℓ1+t2w1,…,tℓ5+t2w5]⊆L[t].B=\mathbb{Q}[t,t\ell_1+t^2w_1,\ldots,t\ell_5+t^2w_5]\subseteq L[t].

The ring BB contains Q[t]\mathbb{Q}[t] and is a Q[t]\mathbb{Q}[t]-submodule of the finite free module L[t]L[t]. It is therefore finite over Q[t]\mathbb{Q}[t]. The fiber B/tBB/tB has only one prime ideal, and its residue field is Q\mathbb{Q}. Indeed, each generator bi=tℓi+t2wi=tai(t)b_i=t\ell_i+t^2w_i=ta_i(t), with ai(t)∈L[t]a_i(t)\in L[t], satisfies a monic equation

bi28+tc1(t)bi27+⋯+t28c28(t)=0,cj(t)∈Q[t],b_i^{28}+tc_1(t)b_i^{27}+\cdots+t^{28}c_{28}(t)=0,\qquad c_j(t)\in\mathbb{Q}[t],

obtained from the characteristic polynomial of multiplication by ai(t)a_i(t) on L[t]L[t]. Thus each bib_i is nilpotent modulo tBtB. The quotient by these generators is QQ, proving the assertion about the fiber.

Let b=(t,b1,…,b5)B\mathfrak b=(t,b_1,\ldots,b_5)B. It follows that B⊗Q[t]Q[t](t)B\otimes_{Q[t]}Q[t]_{(t)} is local and equals BbB_{\mathfrak b}. The map from S0S_0 has image BbB_{\mathfrak b} and kernel p\mathfrak p. In particular, S0/pS_0/\mathfrak p is a one-dimensional domain, and ht⁡p=5\operatorname{ht}\mathfrak p=5. The b\mathfrak b-adic and tt-adic topologies on this finite local algebra are equivalent. Hence

S0/p^=Bb⊗Q[t](t)Q[[t]].\widehat{S_0/\mathfrak p}=B_{\mathfrak b}\otimes_{Q[t]_{(t)}}Q[[t]].

Flatness of Q[[t]]Q[[t]] over Q[t](t)Q[t]_{(t)} embeds this ring in L[[t]]L[[t]]. On the other hand, completion of the quotient presentation identifies it with R/pRR/\mathfrak pR. The resulting map R→L[[t]]R\to L[[t]] is precisely φ\varphi, since it has the same values on all six variables. Thus

pR=P.(45)\mathfrak pR=P. \tag*{(45)}

We next justify (43) by an explicit finite-module kernel. The map

δ:p⟶(S0/p)6,h⟼(∂0h,…,∂5h) mod p\delta:\mathfrak p\longrightarrow(S_0/\mathfrak p)^6,\qquad h\longmapsto(\partial_0h,\ldots,\partial_5h)\mathbin{\bmod}\mathfrak p

is S0S_0-linear by the product rule, and

ker⁡δ=p(2).(46)\ker\delta=\mathfrak p^{(2)}. \tag*{(46)}

This is the differential description of the symbolic square in characteristic zero; in the present setting it also has the following direct proof. If h∈p(2)h\in\mathfrak p^{(2)}, choose s∈S0∖ps\in S_0\setminus\mathfrak p with sh∈p2sh\in\mathfrak p^2. Differentiation and reduction modulo p\mathfrak p show s∂ih∈ps\partial_i h\in\mathfrak p, hence ∂ih∈p\partial_i h\in\mathfrak p for every ii.

Conversely, suppose that h∈ph\in\mathfrak p and all its partial derivatives belong to p\mathfrak p. Write h=a/bh=a/b with a,b∈Q[x0,…,x5]a,b\in\mathbb{Q}[x_0,\ldots,x_5] and b∉nb\notin\mathfrak n. Then aa and all its partial derivatives vanish under the polynomial parametrization into L[t]L[t]. The polynomial change of variables

yi=xi−x0ℓi−x02wi(1≤i≤5)y_i=x_i-x_0\ell_i-x_0^2w_i\qquad(1\leq i\leq5)

therefore shows that aa belongs to (y1,…,y5)2(y_1,\ldots,y_5)^2 in L[x0,…,x5]L[x_0,\ldots,x_5]: both its constant term and all its linear coefficients in the yiy_i vanish. The extension S0→S0⊗QLS_0\to S_0\otimes_{\mathbb Q}L is finite étale. The ideal generated by the yiy_i is a prime contracting to p\mathfrak p. Localizing at that prime gives a faithfully flat local extension of (S0)p(S_0)_{\mathfrak p} with closed fiber a field. As in Proposition 5.2, contraction of its squared maximal ideal gives h∈p2(S0)ph\in\mathfrak p^2(S_0)_{\mathfrak p}. This proves (46).

Since completion is flat, tensoring the kernel description with RR and using (45) gives

p(2)R=ker⁡(P⟶(R/P)6,h⟼(∂ih mod P)i=05).(47)\mathfrak p^{(2)}R=\ker\left(P\longrightarrow(R/P)^6,\quad h\longmapsto(\partial_i h\bmod P)_{i=0}^5\right). \tag*{(47)}

Here the tensor-extended map is indeed formal differentiation modulo PP: if h=∑jajpjh=\sum_j a_jp_j with pj∈pp_j\in\mathfrak p and aj∈Ra_j\in R, the terms pj∂iajp_j\partial_i a_j vanish modulo PP. Proposition 5.2 identifies the kernel on the right with a submodule of P(2)P^{(2)}. The reverse inclusion follows from the same product-rule argument applied to sh∈P2sh\in P^2 with s∉Ps\notin P. This proves (43).

Choose polynomials g1,…,grg_1,\ldots,g_r in Q[x0,…,x5]\mathbb{Q}[x_0,\ldots,x_5] that generate p(2)\mathfrak{p}^{(2)} as an S0S_0-ideal. Such generators are obtained from any finite generating set by clearing denominators that are units in S0S_0. By (43), write the formal witness as

f=∑j=1rajgj,aj∈R.f=\sum_{j=1}^{r}a_jg_j,\qquad a_j\in R.

Choose polynomials ajba_j^b with aj−ajb∈m5a_j-a_j^b\in\mathfrak{m}^5, and set g=∑jajbgjg=\sum_j a_j^b g_j. Then g∈p(2)g\in\mathfrak{p}^{(2)} and g−f∈m5g-f\in\mathfrak{m}^5. In particular, g≡D+W(modm5)g\equiv D+W\pmod{\mathfrak{m}^5}. Proposition 6.2 gives g∉mPg\notin\mathfrak{m}P. If gg belonged to np\mathfrak{n}\mathfrak{p}, its extension to RR would belong to mP\mathfrak{m}P, a contradiction. This proves (44). □

Remark 8.2. Theorem 8.1 proves the existence of a polynomial witness gg with the displayed fourth jet. Its proof approximates the coefficients of a representation of ff in generators of the algebraic symbolic square. It does not identify a degree bound, display the higher coefficients of gg, or assert that a finite truncation of the particular formal series ff is a witness.

Elementary finite certificates

The two finite-field irreducibility checks

Let hˉ=z8−z−1∈F7[z]\bar{h}=z^8-z-1\in\mathbb{F}_7[z]. Repeated modular powering gives

z78≡z(modhˉ),z74−z≡a7(modhˉ),z^{7^8}\equiv z\pmod{\bar{h}},\qquad z^{7^4}-z\equiv a_7\pmod{\bar{h}},

where

a7=z7−3z6+2z5+z4−3z3+2z2.a_7=z^7-3z^6+2z^5+z^4-3z^3+2z^2.

A Bézout certificate is

A7a7+B7hˉ=1in F7[z],A_7a_7+B_7\bar{h}=1\quad\text{in }\mathbb{F}_7[z],

with

A7=3z7+3z5+3z4−z3+2z2+z,A_7=3z^7+3z^5+3z^4-z^3+2z^2+z,
B7=−3z6+2z5−2z4+3z3−z2+z−1.B_7=-3z^6+2z^5-2z^4+3z^3-z^2+z-1.

The finite-field irreducibility criterion proves that hˉ\bar{h} is irreducible: the only prime divisor of 88 is 22.

In F19[z]\mathbb{F}_{19}[z], put

q19=z7+4z6−3z5+7z4+9z3−2z2−8z+5.q_{19}=z^7+4z^6-3z^5+7z^4+9z^3-2z^2-8z+5.

Then

hˉ=(z−4)q19,z197≡z(modq19),\bar{h}=(z-4)q_{19},\qquad z^{19^7}\equiv z\pmod{q_{19}},

and

z19−z≡a19=z5+2z4+z3−z(modq19).z^{19}-z\equiv a_{19}=z^5+2z^4+z^3-z\pmod{q_{19}}.

Here A19a19+B19q19=1A_{19}a_{19}+B_{19}q_{19}=1, where

A19=4z6+2z5−7z4+4z3+5z2−4z+9,A_{19}=4z^6+2z^5-7z^4+4z^3+5z^2-4z+9,
B19=−4z4+6z3+z2−7z+4.B_{19}=-4z^4+6z^3+z^2-7z+4.

Since 77 is prime, these identities prove that q19q_{19} is irreducible. Moreover, q19(4)=9q_{19}(4)=9 in F19\mathbb{F}_{19}, so the displayed factorization of hh is squarefree. All powers in these checks are computed by repeated powering with reduction after each step.

Rational reduction in the coefficient field

For graded reverse lexicographic order with U>VU>V, a Gröbner basis of the ideal (r0,r1)(r_0,r_1) from (9) is

g0=U7−3U3V−U3+4U2V3+6U2V2−UV5−5UV4+V6,g1=U6V−U3+3U2V2+3U2V−UV4−4UV3+V5,g2=U5V2−U6+2U2V+U2−UV3−3UV2+V4,g3=U4V3−2U5V+U2−UV2−2UV+V3,g4=U3V4+U5−3U4V2−UV−U+V2,g5=U2V5+3U4V−4U3V3−U+V,g6=UV6−U4+6U3V2−5U2V4+1,g7=V7−4U3V+10U2V3−6UV5+1.(48)\begin{aligned} g_0 &= U^7-3U^3V-U^3+4U^2V^3+6U^2V^2-UV^5-5UV^4+V^6,\\ g_1 &= U^6V-U^3+3U^2V^2+3U^2V-UV^4-4UV^3+V^5,\\ g_2 &= U^5V^2-U^6+2U^2V+U^2-UV^3-3UV^2+V^4,\\ g_3 &= U^4V^3-2U^5V+U^2-UV^2-2UV+V^3,\\ g_4 &= U^3V^4+U^5-3U^4V^2-UV-U+V^2,\\ g_5 &= U^2V^5+3U^4V-4U^3V^3-U+V,\\ g_6 &= UV^6-U^4+6U^3V^2-5U^2V^4+1,\\ g_7 &= V^7-4U^3V+10U^2V^3-6UV^5+1. \tag*{(48)} \end{aligned}

Their leading monomials are precisely the monomials of total degree seven. Thus Q[U,V]/(r0,r1)\mathbb{Q}[U,V]/(r_0,r_1) has dimension 2828, and its natural surjection onto the degree-2828 field LL is an isomorphism. This also gives the normal form used for the identities involving V0−KV_0-K and the CiC_i.

Reproduction of the finite checks

The supplementary Python program uses exact rational arithmetic and finite-field arithmetic to check the displayed certificates. It checks all 2828 SS-polynomials of (A.1), verifies equality of the ideal they generate with (r0,r1)(r_0,r_1), checks (37) and all 210210 identities (6.7)–(6.8), and verifies the quadratic and cubic Veronese relation-space dimensions. It also checks the normal derivative formulas, the congruence V0≡KV_0\equiv K, and the stated homogeneous lift of V0V_0.

A separate Singular input in the supplementary material reconstructs the resultant and checks its equality with the restriction of WW, together with the normal derivative identities and the nonvanishing of the tangential Hessian determinant in LL. These finite calculations use no floating-point arithmetic and do not replace the all-orders argument in the body of the paper.

Appendix B. Gradient interpolation in higher degrees

The proof of Theorem 1.1 does not require the calculations in this appendix. The purpose here is to describe a direct continuation procedure on the fixed quadratic parametrization and provide additional finite checks of the interpolation mechanism.

For d≥1d\ge1, define the Q\mathbb{Q}-linear map

Jd:Hd⟶L6,E⟼(∂0E(ℓ),…,∂5E(ℓ)).J_d:H_d\longrightarrow L^6,\qquad E\longmapsto\left(\partial_0E(\ell),\ldots,\partial_5E(\ell)\right).

Using the ordered basis (8) in each of the six coordinate blocks, the target has dimension 168168 over Q\mathbb{Q}. Exact rational row reduction gives

dd556677
dim⁡QHd\dim_{\mathbb{Q}} H_d252252462462792792
rank⁡QJd\operatorname{rank}_{\mathbb{Q}} J_d144144160160168168

Table 2.

Here monomial columns are ordered lexicographically by their six-tuples of exponents, starting with the smallest tuple. Each matrix entry is obtained by differentiating a monomial, substituting ℓ\ell, and reducing by the Gröbner basis in Appendix § 9.

Lemma B.1. If J7J_7 is surjective, then JdJ_d is surjective for every d≥7d \ge7.

Proof. For G∈H7G \in H_7, Euler’s identity and ℓ0=1\ell_0=1 give

Jd(x0d−7G)=TdJ7(G),Td(a)=a+d−77(∑j=05ℓjaj)e0.J_d(x_0^{d-7}G)=T_dJ_7(G), \qquad T_d(a)=a+\frac{d-7}{7}\left(\sum_{j=0}^{5}\ell_j a_j\right)e_0.

The LL-linear map TdT_d has determinant d/7≠0d/7 \ne0, so it is bijective. Surjectivity of J7J_7 over Q\mathbb{Q} therefore implies that of JdJ_d. □\square

Proposition B.2. Suppose J7J_7 is surjective and that forms G5∈H5G_5 \in H_5 and G6∈H6G_6 \in H_6 satisfy

φ(∂i(D+W+G5+G6))≡0(modt6)(0≤i≤5).(49)\varphi(\partial_i(D+W+G_5+G_6)) \equiv0 \pmod{t^6} \qquad(0 \le i \le5). \tag*{(49)}

Then there are forms Gn∈HnG_n \in H_n for all n≥7n \ge7 such that

f=D+W+∑n≥5Gnf=D+W+\sum_{n\ge5}G_n

satisfies φ(f)=φ(∂if)=0\varphi(f)=\varphi(\partial_i f)=0 for every ii. Consequently f∈P(2)∖mPf \in P^{(2)}\setminus\mathfrak{m}P.

Proof. Suppose that the gradient of the sum through degree n−1n-1 vanishes modulo tn−1t^{n-1}, where n≥7n \ge7. By Lemma B.1, choose Gn∈HnG_n \in H_n whose image under JnJ_n is the negative of the remaining coefficient of tn−1t^{n-1}. Since

∇Gn(tℓ+t2w)=tn−1Jn(Gn)+O(tn),\nabla G_n(t\ell+t^2w)=t^{n-1}J_n(G_n)+O(t^n),

this kills the next coefficient without changing the preceding ones. The resulting sum converges in the m\mathfrak{m}-adic topology, and all six derivatives belong to PP. The chain rule gives ddtφ(f)=0\frac{d}{dt}\varphi(f)=0; the constant term is zero, so φ(f)=0\varphi(f)=0 as we are in characteristic zero. Propositions 5.2 and 6.2, applied through their proofs, give the last assertion. □\square

The initial forms in (B.1) exist by taking the degree-five and degree-six parts of the series already constructed in Sections 4 and 5. Thus the continuation statement is consistent with the main proof. The rank calculations alone do not prove the initial lifting condition, and no separate explicit coefficient lists for G5,G6G_5,G_6 are needed or asserted here. This appendix is not used as an independent proof of the initial lifting step.

Supplementary material

The accompanying supplementary material consists of a Python program and a Singular input, which carry out the finite checks described in Appendix § 9. The Python program requires only Python 3 with its standard library; the Singular input performs the resultant check in § 3. The recorded runs used Python 3.14.7 and Singular 4.4.1. The checks concern the finite identities; the existence of the full witness is proved by the formal arguments in the text.

References

Shanghai Institute for Mathematics and Interdisciplinary Sciences (SIMIS), Shanghai 200433, China

Email address: [email protected]

Fakultät 8 – Mathematik und Physik, Universität Stuttgart, Stuttgart, Germany

Email address: [email protected]

References

  1. [1][EM97] D. Eisenbud and B. Mazur, Evolutions, symbolic squares, and Fitting ideals, J. Reine Angew. Math. 488 (1997), 189–201. doi:10.1515/crll.1997.488.189.DOI
  2. [2][HH07] M. Hochster and C. Huneke, Fine behavior of symbolic powers of ideals, Illinois J. Math. 51 (2007), no. 1, 171–183. doi:10.1215/ijm/1258735331.DOI
  3. [3][HR98] C. Huneke and J. Ribbe, Symbolic squares in regular local rings, Math. Z. 229 (1998), no. 1, 31–44. doi:10.1007/PL00004649.DOI
  4. [4][Mor15] A. A. More, A note on the Eisenbud–Mazur conjecture, J. Pure Appl. Algebra 219 (2015), no. 7, 274–275. doi:10.1016/j.jpaa.2014.09.025.DOI
  5. [5][Mur23] T. Murayama, Uniform bounds on symbolic powers in regular rings, arXiv:2111.06049v5 [math.AC], 2023.arxiv.org/abs/2111.06049

Paper details

vlnc:W_yyc6af9c
Version
v1 · latest
Published
Added
License
CC BY 4.0
Usage

views · 7

downloads · 2

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

OpenAI ChatGPT GPT-6 Astra, Anthropic Claude Fable 5.1

View statement
Looking for related papers…
Contents