WorksMathematics
A counterexample to the Eisenbud–Mazur conjecture
Abstract
We construct a height-five prime ideal in the six-variable formal power series ring over such that , thus providing a prime-ideal counterexample to the Eisenbud–Mazur conjecture (with being the coefficient field). The ideal is the kernel of an explicit quadratic parametrization over a number field of degree over . A witness has order three, with leading term the determinant of a symmetric matrix. We construct the witness by eliminating critical values to all orders and then straightening the resulting formal curve. Interpolation at the nodes of an arrangement of eight lines supplies the required lifting maps. A rational linear functional on third and fourth jets certifies nonmembership in . The construction also yields an unmixed radical ideal with smooth branches over , and provides a counterexample to the more general form of the Eisenbud–Mazur conjecture stated by More (with being the coefficient field). Finally, the prime ideal counterexample descends to the localized polynomial ring , 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
for every prime ideal in a regular local ring containing a field of characteristic zero. Here is the symbolic square of . Equivalently, it predicts that no element of the symbolic square can be a minimal generator of . 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 . The quotient by is a one-dimensional complete local domain, and its normalization has a residue-field extension of of degree 28. After extending coefficients to , 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 ; 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 [4 Theorem 2.1]. A related uniform bound is , where is the height of : 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: if and only if and all its partial derivatives lie in [1 Theorem 3]. So (1) is a statement about critical loci: a witness is a hypersurface that is singular along and is nevertheless a minimal generator of . Second, if is stable under a derivation with positive integer weights, the Euler identity places every such in [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 the differential description breaks down, and [1] exhibits monomial curves in characteristics 2 and 3 that violate (1).
We set
The first two nonzero homogeneous terms of our witness are the cubic
and the quartic
The significance of the quartic is explained after Theorem 1.1; see also § 3 and § 4.
Theorem 1.1. There is an explicitly defined continuous -algebra homomorphism
where is a number field of degree 28 over , with for , such that is a height-five prime ideal and there exists satisfying
The normalization of is , whereas the residue field of is .
The field and all coefficients of are specified by rational polynomial formulas below. The witness 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 . The projectivized singular locus of is the quadratic Veronese surface: the partial derivatives of are the minors, which generate the prime ideal of the cone over this surface, so that . The Laplace expansion, however, writes as a combination of the minors with coefficients in , so and itself is no counterexample. We therefore replace 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 can be a minimal generator; what has to be arranged is that its gradient vanishes along the arc, and the quartic supplies the critical point that makes this possible. On its affine chart
the restriction of is the product of the eight lines , where runs through the roots of . Their 28 nodes form a single Galois orbit. Evaluation at one node identifies the polynomials in of degree at most six with . Consequently, evaluation at the corresponding point of the Veronese surface maps homogeneous cubics in the six variables onto , 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 . This permits a formal coordinate change with identity two-jet that replaces the critical arc by the prescribed quadratic parametrization while preserving modulo . 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 can be checked using only the fourth jet. We determine the image of modulo and display a rational linear functional supported in degrees three and four after quotienting by . It annihilates the image of and takes the value 1 on . 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 to 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
Throughout the paper, orders and homogeneous components refer to ordinary total degree in . We write for the space of homogeneous forms of degree . Let
choose two distinct roots in an algebraic closure of , and put
Lemma 2.1. The extension has degree 28. Its embeddings into a splitting field of are indexed by the unordered pairs of distinct roots of .
Proof. The polynomial is irreducible modulo 7. Modulo 19 it factors as
where the degree-seven factor is irreducible. Both reductions are squarefree; exact certificates are given in Appendix A. Consequently the Galois group of 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 determines as the roots of . Its orbit thus has size 28, which is . □
For each root of , consider the line
The lines corresponding to and meet at . No three are concurrent, since a monic quadratic cannot have three distinct roots.
Lemma 2.2. Evaluation at gives an isomorphism of -vector spaces
and with:
the evaluation map , , is surjective for every .
Proof. Suppose that vanishes at . Lemma 2.1 implies that it vanishes at all 28 intersections. Each line contains seven distinct intersections, so the restriction of to that line is zero. Over the splitting field, the product of the eight distinct line equations therefore divides . Since , we obtain . Both vector spaces have dimension 28, proving the first assertion.
Every monomial with is a product of at most three elements of . Padding with powers of gives a homogeneous cubic restricting to this monomial under
Multiplication by proves the assertion for all . □
We use the ordered basis
where and then increase, and where the entries are evaluated at when regarded as elements of . For exact computations, one can instead use the relations
Indeed, division of by leaves remainder . The quotient then identifies with ; 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 is an isomorphism.
The explicit formal curve
Retain the field from Lemma 2.1 and its ordered basis of monomials of degree at most six. Define the resultant:
Explicitly:
If ranges over the eight roots of in a splitting field, then
We use the affine coordinates
For a homogeneous polynomial in the variables , write for its substitution under (11). Direct expansion gives
The three normal derivatives of on , namely , and similarly for and , are the following polynomials:
Set
Lemma 3.1. At one has
and is invertible.
Proof. The point is the intersection of the two lines indexed by and . No third line passes through it. Let be the product of the other six line equations, evaluated at this point. Then , and differentiating the product of the eight line equations gives
The same product description gives the three stated vanishings of . Finally, expansion of using (13) and then reduction by the two relations (9) gives
The displayed reduction basis in the finite-certificate appendix makes this a rational polynomial identity; the supplementary verifier checks it independently. This shows that vanishes at . □
All derivatives in the next definition are taken as polynomial derivatives in before evaluation at . Define by
Put
These formulas specify a continuous -algebra homomorphism from to , i.e. an arc:
(Here is the -th coordinate of .) In particular, . Geometrically corresponds to a formal curve in the formal spectrum of .
Lemma 3.2. For every ,
The image is a one-dimensional complete local domain, finite over , with residue field . Its fraction field is , and its normalization is . Consequently, is a height-five prime ideal and .
Proof. By Lemma 2.2, choose homogeneous cubics such that form a -basis of . Relative to this basis, the elements form the columns of a matrix in whose reduction modulo is invertible. They therefore form a -basis of . Since , every element of lifts to an element of . Multiplication by proves the reverse inclusion in (18); the forward inclusion follows from .
The image contains and is a -submodule of the finite free module . It is therefore finite over . It is a domain of dimension one. As a quotient of the complete local ring , it is complete and local. Reduction modulo takes onto ; its kernel is the image of , so its residue field is . The inclusion implies that . The ring is finite over and integrally closed in this field, hence is the normalization of .
It follows that . The six coordinates of are linearly independent over by Lemma 2.2. The coefficient of in therefore excludes a nonzero linear initial form for any . The constant coefficient also vanishes, so .
Eliminating critical values to all orders
Consider the homogeneous cubic
Its restriction to under (11) is exactly . Set .
Proposition 4.1. There exist homogeneous forms for all and a continuous -algebra homomorphism , i.e. an arc, such that with:
one has:
the critical condition:
and
Proof. We first associate to every (tentative) an arc that satisfies the condition:
First, make the following substitution (the affine cone version of (11)):
into , and then divide by ; the resulting formal series is then:
We work over (and thus ). The Hessian of in the three variables is invertible. The formal implicit function theorem then gives unique critical solutions in satisfying:
Here are the polynomials in as before; the corresponding critical equations are:
Let be the result of substituting these three solutions into . A calculation through degree two in gives
Put . Lemma 3.1 and the formal implicit function theorem give unique series with constant terms satisfying the critical equations:
Their first terms are
by (15).
For abbreviations we write:
for the elements of obtained by substituting into the variables .
Substituting into the variables , we then have the formulas
which then defines an arc .
Now (in the affine chart) the change of coordinate map
has determinant-one Jacobian. The five critical equations that the five critical solutions satisfy, thus give:
Note also that equations (24) and (26) give precisely the coefficients in (16). Thus every such satisfies (21).
Now define the critical value of by:
We thus have by definition:
The vanishings in Lemma 3.1 give:
and hence (25) then implies
Next we claim, for a homogeneous , , that
For this, introduce a parameter and carry out the two recursive implicit constructions for . Each coefficient of the resulting arc lies in : 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 . Since is fixed and the five spatial derivatives vanish, it gives
Integrating from to 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 . Inductively suppose that has been chosen and that its critical value has no term of degree less than . Let be the coefficient of . By Lemma 2.2, choose with . Equation (29) shows that kills the coefficient of while preserving all previous vanishings.
Now adding a form of degree changes and its normal equations by . The fixed invertible normal Hessian then shows that the normal critical solutions, as functions of , change by . After reduction and division by , the tangential equations change by . Their fixed invertible constant Hessian shows that change by . Since the normal solutions and their derivatives are divisible by , their evaluations on the final tangential arc change by . The original coordinates therefore change by . These orders tend to infinity with .
It follows that converges -adically to the series with the shape in (20), and the arcs converge coefficientwise to an arc satisfying (21). Substitution and formal differentiation are continuous for these filtrations, so the induction and (27) give
Finally to complete the proof of (22), it remains to show that . Differentiate with respect to . The chain rule and give
which thus completes the proof of (22).
Straightening and symbolic-square membership
Lemma 5.1. Let be any arc that satisfies (21). There is a continuous -algebra automorphism of such that
Proof. Each difference belongs to . By (18), choose whose image under is this difference. Define for and . This substitution has identity linear part, so recursive solution of its homogeneous terms gives a continuous inverse. Its composite with agrees with on the six variables and hence on by continuity.
Apply this lemma to the arc of Proposition 4.1, and put
Changing variables by terms of order at least three changes a cubic only in degrees at least five. Therefore
The chain rule and imply
Proposition 5.2. The series in (30) belongs to . More generally, if satisfies for , then .
Proof. Let and set
The same formulas that define as before then give a surjective homomorphism with kernel , and . Use and as formal coordinates on . In terms of these new coordinates, we have , and then is the ideal of generated by for , and hence is the ideal of generated by for . One then sees immediately, by using Taylor expansion for example (with respect to the original coordinates , with being regarded as fixed), that if satisfies for , then . Since is finite and separable, is finite étale over . The local map is faithfully flat. Its closed fiber is a localization of a finite étale algebra over , hence a field. Consequently,
Faithful flatness therefore gives
Intersecting with proves . Finally the particular series satisfies the hypotheses by (32). □
A finite certificate for nonmembership
Put
A bar denotes the image under the substitution (here we identify with ). We shall construct a -linear functional such that
Together with (31), these identities prove the required nonmembership in .
Consider the six quadrics:
with their accompanying homogeneous cubic terms given by:
Lemma 6.1. The image of in is the rational span of the classes of
Proof. Let denote substitution of into homogeneous forms. The degree-two kernel of is spanned by the six , and its degree-three kernel is spanned by their linear multiples. These statements can be seen by homogenizing the substitution to the map
The six quadrics in (34) generate all relations among these degree-two monomials: any two products representing the same monomial in are related by changing the pairing of four factors. The quotient dimensions in degrees two and three are therefore 15 and 28, respectively. Since consists of polynomials of degree at most , Lemma 2.2 shows that evaluation at has these same kernels for .
By the definition of as given in (17), the coefficient of in is
The negatives of these six coefficients, in order, are
where are evaluated at . The terms containing cancel. Substitution into the displayed , followed by reduction using (9), gives exactly the values in (37). Consequently,
For , Lemma 3.2 gives with . The coefficient of in gives for some . By (38), the cubic evaluates to zero at . It is therefore a linear combination of the , proving one inclusion.
Conversely, every expression in (36) maps into . Lemma 3.2 gives , so its image can be canceled by an element of . Its class modulo thus belongs to the image of .
Now define the functional on monomials by Table 1, and set it to zero on every monomial not listed. It extends to all of by taking this finite linear combination of coefficients; in particular, .
| Degree-three tuple | Value | Degree-four tuple | Value |
Table 1. All nonzero values of . An exponent tuple denotes .
Proposition 6.2. The functional satisfies (33). Every series consequently lies outside .
Proof. The displayed coefficients give the rational identities
These are 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 , 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 , so this leaves two degrees of freedom to define the functional satisfying (39)–(40) (these two remaining degrees of freedom are used to make and as shown below; incidentally this also explains why is defined this way).
For , choose such that . The terms of through degree four depend only on the classes of the modulo . Lemma 6.1 and (39)–(40) therefore give . This proves vanishing of on the whole ideal .
and
Table 1 gives and . Thus whenever . If , then , a contradiction.
Proof of Theorem 1.1. The field and 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 given in equation (30), with . Proposition 5.2 gives , and Proposition 6.2 gives .
Extension of the coefficient field
For an ideal in a Noetherian ring, recall the symbolic square:
The prime ideal of in the previous constructions acquires a particularly simple branch description after extension of the coefficient field.
Proposition 7.1. Let , let , and put . Then is an unmixed radical ideal, and we have the primary decomposition:
These 28 ideals are the distinct smooth branch primes of , and the series of equation (30) satisfies
Proof. By Lemma 3.2, the inclusion is an inclusion of finite -modules. The maximal-ideal topology on agrees with its -adic topology: the quotient is a finite-dimensional local -algebra. Completed extension of the coefficient field consequently gives
For completeness, this identification can be seen by reducing the finite -module modulo , extending coefficients to , 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 .
The extension is flat, since its target is torsion-free over the discrete valuation ring . It follows that
The image contains times the product on the right. The coordinate maps have the kernels displayed in (41); their intersection is therefore . Each quotient is , and distinct embeddings give distinct pairs , hence distinct tangent directions. Thus these are precisely the minimal primes of .
The identities persist on every branch. Taylor expansion in the five transverse branch coordinates therefore places in the square of each branch ideal, and hence in . Extend the functional of Table 1 -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 modulo and takes value on , proving the nonmembership. □
Corollary 7.2. There is a formal power series without constant term such that
Proof. Take the series given by equation (30). The radical ideal of Proposition 7.1 contains and all its partial derivatives, so the displayed radical is contained in . The assertion follows from .
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 being the coefficient field).
A counterexample in a localized polynomial ring
The formulas that define also define an algebraic curve. The construction of 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
where the same notation denotes the maximal ideal after localization. Let be the kernel of the polynomial homomorphism
and put .
Theorem 8.1. The ideal is a height-five prime of . Its extension to the completion is , and
There exists a polynomial such that
In particular, the counterexample exists in a regular local ring essentially of finite type over .
Proof. We first identify the completed prime. Write
The ring contains and is a -submodule of the finite free module . It is therefore finite over . The fiber has only one prime ideal, and its residue field is . Indeed, each generator , with , satisfies a monic equation
obtained from the characteristic polynomial of multiplication by on . Thus each is nilpotent modulo . The quotient by these generators is , proving the assertion about the fiber.
Let . It follows that is local and equals . The map from has image and kernel . In particular, is a one-dimensional domain, and . The -adic and -adic topologies on this finite local algebra are equivalent. Hence
Flatness of over embeds this ring in . On the other hand, completion of the quotient presentation identifies it with . The resulting map is precisely , since it has the same values on all six variables. Thus
We next justify (43) by an explicit finite-module kernel. The map
is -linear by the product rule, and
This is the differential description of the symbolic square in characteristic zero; in the present setting it also has the following direct proof. If , choose with . Differentiation and reduction modulo show , hence for every .
Conversely, suppose that and all its partial derivatives belong to . Write with and . Then and all its partial derivatives vanish under the polynomial parametrization into . The polynomial change of variables
therefore shows that belongs to in : both its constant term and all its linear coefficients in the vanish. The extension is finite étale. The ideal generated by the is a prime contracting to . Localizing at that prime gives a faithfully flat local extension of with closed fiber a field. As in Proposition 5.2, contraction of its squared maximal ideal gives . This proves (46).
Since completion is flat, tensoring the kernel description with and using (45) gives
Here the tensor-extended map is indeed formal differentiation modulo : if with and , the terms vanish modulo . Proposition 5.2 identifies the kernel on the right with a submodule of . The reverse inclusion follows from the same product-rule argument applied to with . This proves (43).
Choose polynomials in that generate as an -ideal. Such generators are obtained from any finite generating set by clearing denominators that are units in . By (43), write the formal witness as
Choose polynomials with , and set . Then and . In particular, . Proposition 6.2 gives . If belonged to , its extension to would belong to , a contradiction. This proves (44). □
Remark 8.2. Theorem 8.1 proves the existence of a polynomial witness with the displayed fourth jet. Its proof approximates the coefficients of a representation of in generators of the algebraic symbolic square. It does not identify a degree bound, display the higher coefficients of , or assert that a finite truncation of the particular formal series is a witness.
Elementary finite certificates
The two finite-field irreducibility checks
Let . Repeated modular powering gives
where
A Bézout certificate is
with
The finite-field irreducibility criterion proves that is irreducible: the only prime divisor of is .
In , put
Then
and
Here , where
Since is prime, these identities prove that is irreducible. Moreover, in , so the displayed factorization of 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 , a Gröbner basis of the ideal from (9) is
Their leading monomials are precisely the monomials of total degree seven. Thus has dimension , and its natural surjection onto the degree- field is an isomorphism. This also gives the normal form used for the identities involving and the .
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 -polynomials of (A.1), verifies equality of the ideal they generate with , checks (37) and all identities (6.7)–(6.8), and verifies the quadratic and cubic Veronese relation-space dimensions. It also checks the normal derivative formulas, the congruence , and the stated homogeneous lift of .
A separate Singular input in the supplementary material reconstructs the resultant and checks its equality with the restriction of , together with the normal derivative identities and the nonvanishing of the tangential Hessian determinant in . 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 , define the -linear map
Using the ordered basis (8) in each of the six coordinate blocks, the target has dimension over . Exact rational row reduction gives
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 , and reducing by the Gröbner basis in Appendix § 9.
Lemma B.1. If is surjective, then is surjective for every .
Proof. For , Euler’s identity and give
The -linear map has determinant , so it is bijective. Surjectivity of over therefore implies that of .
Proposition B.2. Suppose is surjective and that forms and satisfy
Then there are forms for all such that
satisfies for every . Consequently .
Proof. Suppose that the gradient of the sum through degree vanishes modulo , where . By Lemma B.1, choose whose image under is the negative of the remaining coefficient of . Since
this kills the next coefficient without changing the preceding ones. The resulting sum converges in the -adic topology, and all six derivatives belong to . The chain rule gives ; the constant term is zero, so as we are in characteristic zero. Propositions 5.2 and 6.2, applied through their proofs, give the last assertion.
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 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][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][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][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][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][Mur23] T. Murayama, Uniform bounds on symbolic powers in regular rings, arXiv:2111.06049v5 [math.AC], 2023.arxiv.org/abs/2111.06049