WorksMathematics
Power loss for the local Mizohata–Takeuchi estimate on smooth hypersurfaces
Abstract
We prove power-type lower bounds for the local Mizohata–Takeuchi constants for hypersurfaces with whose second fundamental form is nonzero at some point. This result, in particular, disproves the so-called local Mizohata–Takeuchi conjectures for the paraboloid and the sphere and, in turn, presents an obstruction to a general weighted approach to the theory of Fourier restriction of those classical surfaces.
Introduction
Background. In the late 1970s, Stein [23] raised the question of whether weighted estimates for Fourier multiplier operators, notably the ball and Bochner–Riesz multipliers, could be controlled by Kakeya or Nikodym-type maximal operators (see [26] for a more detailed exposition about this question). In the context of Fourier restriction theory, this leads to what is commonly known as Stein’s conjecture, which states that weighted norms of Fourier extension operators are controlled by maximal X-ray averages of the weight along lines normal to the underlying hypersurface.
More precisely, let be a hypersurface in , and define the Fourier extension operator
where is the surface measure of and . For a locally integrable weight and , define
Stein’s conjecture reads as follows:
Conjecture 1.1 (Local Stein conjecture). For every and , we have
for all and nonnegative weights supported in the ball of radius centred at the origin, where denotes a unit normal to at .
An immediate consequence of (1) would be the Mizohata–Takeuchi conjecture:
Conjecture 1.2 (Local Mizohata–Takeuchi conjecture). For every and , we have
for all and nonnegative weights supported in .
The Mizohata–Takeuchi conjecture 1.2 has its origins in the study of well-posedness for Schrödinger-type evolution equations in [18] and [25]. In the 1990s, Barceló, Ruiz and Vega [5] studied a similar problem by establishing certain weighted estimates for the Helmholtz equation, and obtained applications to both Fourier restriction and the well-posedness of Schrödinger equations with time-dependent drift terms. The formulation (2) stems from [5].
The global versions of Conjectures 1.1 and 1.2 state that
and
for all and weights . These global variants were first disproved by one of the present authors in [8]. See also [2] for a phase-space formulation of (3) and (4), and [3] for a different set of counterexamples to (3) and (4).
Recent positive results include the global estimate for uniformly convex planar curves with semialgebraic indicator weights of bounded complexity due to Hao, Wang and one of the present authors [14], and the local bound for the hyperbolic paraboloid in due to Kim and Wang [15].
A further motivation for Stein’s conjecture is its potential to establish a direct connection between two of the central problems in harmonic analysis: the Kakeya maximal conjecture and the Fourier restriction conjecture1. The latter is known to imply the former (see [17]), but combining Conjecture 1.1 with the conjectured bound for the Kakeya maximal operator would give
which is the local endpoint formulation of the Fourier restriction conjecture for smooth compact hypersurfaces with nonvanishing Gaussian curvature; see [4]. Thus, Stein’s conjecture would provide a mechanism for transferring geometric control of concentration along tubes into optimal Fourier extension estimates.
Failure of Conjecture 1.2
Let the -local constant be defined by
The local Mizohata–Takeuchi conjecture 1.2 can then be restated as
Guth [12] noted that axiomatic wave-packet decomposition alone cannot prove Conjecture 1.2; this was fleshed out by Carbery, Iliopoulou and Wang [10], where the bound was established for strictly convex hypersurfaces with nonvanishing Gaussian curvature, alongside its optimality using only wave-packet axioms. Two of the present authors [9] subsequently showed that there exists a set of hypersurfaces for which the reverse inequality holds (see also [11]). Thus, for a generic hypersurface, the extent to which the Mizohata–Takeuchi conjecture holds has been settled entirely, up to the endpoint. On the other hand, both [9] and [11] deal with random surfaces, and thus the possibility that Conjecture 1.2 may hold for -surfaces (and in particular, the paraboloid and the sphere) remained open; see the recent discussion in [27 Sec. 6].
Our goal is to prove that Conjecture 1.2 fails for every sufficiently smooth connected hypersurface that is not contained in a hyperplane. Recall that a hypersurface is of class for some if, near each point and after a rigid change of coordinates, it becomes the graph of a function whose Hessian is Hölder continuous with exponent . Thus, holds for all in the same chart, where the implicit constant is allowed to depend on the chart.
The main result of this paper is the following:
Theorem 1.3. Take and suppose that is a hypersurface whose second fundamental form is nonzero at some point. Then for all we have
where depends only on the smoothness parameter (and is, in particular, independent of the dimension and the curvature of ).
The proof brings together ideas from harmonic analysis and algebraic number theory. We first reduce Theorem 1.3 to the construction of finite sets with large parabolic additive energy and controlled line occupancy. The next step is to use a tensor-power argument to amplify the gain from the counterexample in [8] to an exponential one in very high dimensions. The last challenge is to transfer this gain back to the real line without destroying the underlying additive relations or the required separation properties, which can be done with tools from algebraic number theory. The resulting constructions yield the desired power loss for the parabola, which is then transferred to general curved hypersurfaces.
Outline of the paper. This paper is organised as follows. In Section 2, we present an outline of the argument. In Section 3, we reduce Theorem 1.3 to a statement about additive energy of certain well-structured configurations on the parabola. Section 4 uses techniques from algebraic number theory to produce power-large energy of these configurations, which are then applied to complete the proof of Theorem 1.3 in Section 5. Appendix A records the standard number-theoretic facts which we use in Section 4.
Notation. For quantities , we write or if for some absolute constant ; means . We write if for which is sufficiently small. If the implicit constants depend on some auxiliary parameter , we write, for example . The notation will be introduced in Section 4. We also write for the cardinality of a finite set .
Acknowledgements. The authors acknowledge the support of their individual fellowships and grants.
H.C. is supported by NSF GRFP.
I.F. was supported by the NSF RTG grant No. DMS-2230900.
V.K. was supported in part by the Croatian Science Foundation under the project HRZZ-IP-2022-10-5116 (FANAP) and in part by the European Union – NextGenerationEU through the National Recovery and Resilience Plan 2021–2026, via an institutional grant of the University of Zagreb Faculty of Science, IK IA 1.1.3, Impact4Math.
I.O. was supported by his EPSRC Postdoctoral Fellowship UKRI3285. Part of this work was conducted while he was in residence at the Institut Mittag-Leffler in Djursholm, Sweden, during the autumn semester of 2026.
D.Z. was supported by the Simons Dissertation Fellowship.
R.Z. is supported by NSF CAREER DMS-2143989 and Sloan Research Fellowship. He would like to thank Alex Cohen for helpful discussions.
Outline of the argument
The outline of the proof of Theorem 1.3 is organised into three parts: the analytic reduction in Steps (1)–(2), the combinatorial construction in Steps (3)–(4), and its arithmetic realisation in Steps (5)–(7). We present the ideas in the order that motivates them.
(1) Reduction to the parabolic case. The non-vanishing curvature assumption on ensures that, for some 2-plane , the curve is well approximated by a parabola (Lemma 3.3). Our function and weight will essentially lie on this one-dimensional slice, extended in the remaining dimensions isotropically. This allows us to reduce power loss on to power loss on the parabola ; the Hölder assumption ensures that the approximation is close enough to maintain power loss.
(2) Reduction to parabolic energy bounds. We perform a further reduction to show that the Mizohata–Takeuchi conjecture on implies certain energy bounds on “well-separated” sets , (Proposition 3.1). More precisely, it is enough to make the quotient
grow as a positive power of , where is the separation scale, is the parabolic additive energy, and is the “maximal strip occupancy” of the set . The energy gives a lower bound for the weighted extension norm (Lemma 3.4), while strip occupancy gives an upper bound for the X-ray norm of (Lemma 3.5). Our goal is then reduced to finding such sets , .
(3) Finding a seed configuration. We begin by finding “seed” sets and whose energy ratio exceeds any prescribed constant (see (4.4)); these seeds will be combinatorial in nature, based on the counterexamples from [8]. This will translate to a disproof of the global Mizohata–Takeuchi conjecture (1.4), but will not by itself give power loss.
(4) Amplification by tensor products. Next, we consider the large product sets , ; we refer to this product configuration as Product (4.5). Then one can see that Product witnesses power loss for the desired energy ratio, with the relevant gain becoming exponential in the number of tensor factors, but does so in instead of . This forces us to deal with two problems:
(i) The coordinate projection maps are highly non-injective, and directly projecting Product will lose all of the gain.
(ii) The natural way to deal with (i) would be to slightly rotate our configuration so that the coordinate maps do become injective on Product. This introduces the more serious issue that projection will no longer be a ring homomorphism, and in particular will not preserve the energy equation; moreover, we will have very little control on the point separation after projection.
(5) Introducing an auxiliary lattice. To deal with (ii), we note that, if we could find Product inside the ring of integers of a totally real number field , embedded in via its Minkowski embedding, then both difficulties (i) and (ii) above would be resolved; we refer to such a high-dimensional lattice as Structure. Indeed, in such a case all coordinate projections are injective ring homomorphisms (see the discussion following (4.11)). However, this is unlikely to work directly, since Product is discrete (and quite sparse), so there is a priori no reason to expect Product to be large. (6) Thickening and intersecting. We next perform a parabolic rescaling, enlarging Product to be at a scale much larger than Structure, and then thickening the points of this rescaled copy to be rectangular regions whose sizes are large relative to Structure but small relative to the rescaling; call the resulting object Product (16). Assuming the scales are chosen correctly, all of the thickenings in Product remain well separated (Lemma 4.8), but now we can expect that Structure will intersect these large rectangular regions (see Figure 2). To make this precise, we require quantitative dimension-independent control on intersections of Structure with axis-aligned boxes (Lemma 4.3).

Figure 2. A two-dimensional schematic of the parabolic rescaling and thickening on the side. The hollow circles indicate the rescaled seed centres . Each is thickened to a square of side length . Since , these rectangles remain well separated. The oblique lattice is shown in blue, while the lattice points lying in are shown in black.
(7) Projection back to the real line. Once all of this is arranged, we choose the parameters in the following order. Fix the root discriminant , then choose the seed size and the rescaling parameter , independently of . We then let tend to infinity through the powers of supplied by Lemma 4.2. The intersection of Product with Structure gives us sets and so that the ratio (5) has size at least inside of when lines are restricted to being non-vertical in , where is some quantity depending on but not (Lemma 4.9). Choosing sufficiently large relative to makes this quotient at least (see the proof of Proposition 4.1). We then project (using field homomorphisms) and parabolically rescale to obtain and without changing their cardinalities or parabolic energy (see (20) and the discussion following it). The required separation and thin-strip occupancy follow from the arithmetic gap estimates (Lemmas 4.14 and 4.15). The separation parameter is , with independent of , while the corresponding spatial radius is (Proposition 3.1). Thus the exponential gain becomes a positive power of .
Here are the rough sizes of the involved quantities, once is fixed:
A transfer argument
In this section, our goal is to reduce the local Mizohata–Takeuchi conjecture 1.2 on an arbitrary hypersurface to an energy estimate on the parabola ; this will correspond to Steps (1)–(2) of the outline.
Let , and define for any ring and sets , the parabolic additive energy
We will typically suppress the dependence on the ring from the notation.
Proposition 3.1. Suppose and . Suppose and are finite nonempty sets so that is -separated, and so that
for every and interval of length . Then if and is a hypersurface whose second fundamental form is nonzero at some point, we have
for some .
Remark 3.2. Notice that the strip occupancy of the set is measured with respect to strips with ‘vertical width’ , i.e. ones intersecting the vertical lines in an interval of length . This is in contrast to the X-ray transform in Conjecture 1.2 which more naturally corresponds to strips of (actual) width . This distinction is due to the parabolic rescaling step in the proof below, and allows us to weaken the restrictions on the desired set .
From now on, implicit constants will be permitted to depend on . In the local coordinates of Lemma 3.3, write , where the Jacobian satisfies . Moreover, we identify a function on the parameter domain with the function on , so that
with . We begin by fixing some parameters (the implicit constants below will be chosen later, but they are not important at this stage):
Note that we have .
We begin with the following lemma, which follows from the fact that is with somewhere nonvanishing curvature.
Lemma 3.3. Up to an element of , we may locally describe in the form
where , with , , , and .
Note that in the coordinates of Lemma 3.3, if we let , then is approximated by the standard parabola. This change of coordinates will affect all estimates by a bounded multiplicative factor depending only on the surface (which we will absorb into our asymptotic notation).
Next, we set and . For each , we set . We then set
Since is -separated, the set is -separated, so we immediately obtain
Next, we choose the weights. Define
where
We see that is essentially an isotropic thickening of a 1-dimensional function corresponding to the slice , and similarly is an isotropic thickening of a 2-dimensional weight. Note that is continuous and supported in .
Lemma 3.4. We have
Proof. Developing gives
where for the kernel is given by
Since , restricting the sum above to the tuples that contribute to the parabolic energy gives
Fix in the summand of (8). Write , and , and notice that direct substitution and Lemma 3.3 give
and so by appropriately choosing the implicit constants in (3.1) we can make all of these values , in which case we obtain on , hence
Finally, we must bound the X-ray transform of .
Lemma 3.5. For any affine line , we have
Proof. We begin with a one-dimensional claim: if and are arbitrary frequencies so that every interval of length contains at most such frequencies, then
for any coefficients of absolute value 1. Indeed, this follows from expanding the square and applying the decay of alongside appropriate truncations.
Now, let , where . We split into two cases.
Case 1: . Making a change of variables yields the bound
where
Note that the implicit constant is uniform in our range of . Then lying in an interval of length is equivalent to lying in an interval of length , for , and hence we may apply the bound (9) to obtain
Case 2: . Then some coordinate of , say , has size , and so we can make a change of variables using the trivial bound on the associated exponential sum, yielding
where the final bound comes from the observation .
:::
Proof of Proposition 3.1. Choose and as above. Then combining (3.2) with Lemmas 3.4 and 3.5 yields
as desired. □
Arithmetic amplification
In this section, we provide a construction of power-large energy using techniques from algebraic number theory; this will comprise Steps (3)–(6) of the outline. This argument follows ideas of the disproofs of the unit-distance conjecture [1] and the sum-product conjecture [7], as presented in Bloom [6]. Our goal is to prove the following proposition.
Proposition 4.1. There exists some absolute so that the following holds. For all , there exist sets and satisfying the following conditions:
(i) is -separated.
(ii) For all and all intervals of length , we have
for some .
(iii) We have
The geometry of conditions (i)–(ii) is illustrated in Figure 1.

Figure 1. Schematic geometry in Proposition 4.1, with . The shaded strip has vertical width ; the occupancy bound holds for every and every interval of this length.
Structure
To start, we record the quantitative properties we need of Structure. Throughout this subsection, let be a totally real number field of degree , with embeddings (see Appendix A for other standard number-theoretic notation). We will use the elementary norm bound from Lemma A.1, proved in Appendix A.
We require the following arithmetic input, which supplies us with an infinite family of totally real fields of bounded root discriminant. This construction is due originally to Martinet [16].
Lemma 4.2. There is some so that, for all , there exists a totally real field of degree and root discriminant .
Even though this will not affect in any way the results of this paper, it is perhaps helpful to think of as an explicit number. A concrete choice coming from Martinet’s construction is
To simplify notation, we employ the following convention: once the root discriminant has been fixed as in Lemma 4.2, write for for some which is allowed to depend on , but not on the parameters , , , (some of which have not yet been introduced). Also write if . We will always be able to take arbitrarily large in relation to , and so this loss will always turn out to be tolerable.
Lemma 4.3. Let be a totally real number field of degree and root discriminant , and let . Then
for any axis-aligned rectangle , with and a sufficiently large constant depending on .
Proof. The upper bound follows easily from a packing estimate, since distinct points in are 1-separated in the -metric, and . Indeed, for two distinct points with , , we have . Hence its norm and
which yields . This way, the cubes
are pairwise disjoint. Since , their union satisfies
Therefore . Since , this gives the desired upper bound. It thus suffices to show the lower bound.
To start, let be the dual lattice of . Given any , we see that . Thus, if is nonzero, multiplication by sends a lattice of covolume to one of covolume at least (see (A.1)), and in particular the determinant of this transformation is at least ; but this determinant is precisely , and so
for all nonzero.
Let be a fixed bump function supported on , with . Also assume we have Fourier decay . One can for example take . Choose a rectangle , let , and set . By Poisson summation, we have
Note that , and so it suffices to bound the remainder
Fix integers , and consider the dyadic regions
Since , we have for the estimate
Now, we estimate the intersections . For all , we have . By pigeonholing in the first coordinate, we can moreover find some pair for which
Applying (10) for and (11) and reorganising gives
Thus, denoting , we can estimate
The sum converges and so we obtain an overall estimate . Recall that the main term contribution is and by the assumption . Taking larger than the constant in the notation , we conclude that the main term dominates and . □
Remark 4.4. By carefully tracking through the constants in Lemma 4.3, one can show that the implicit constant and constants in the notation are absolute (meaning independent of ), and in fact we have the bound
Remark 4.5. Lemma 4.3 can be viewed as a quantitative version of the well-known fact that the orbit of under the diagonal group is compact in the space of lattices in . See [22] and the references therein for details on this perspective.
Product 0 and Product
Next, we produce our base seed, based on [8].
Lemma 4.6. For every integer , there exist distinct integers such that the points
are pairwise distinct and no three of them are collinear.
Proof. For pairwise distinct subsets , consider the polynomial
Writing , multilinearity in the first two rows gives
We claim that is not identically zero. Indeed, if that were the case, all would vanish. Since the subsets are pairwise distinct, at least one vector is neither nor . Up to relabelling the subsets , we may assume without loss of generality that . Computing gives
Since the first factor is , we obtain for every , which is a contradiction since .
We now choose the so that all these determinants are simultaneously nonzero. Fix a total order on the subsets of , and set
Every factor is a nonzero polynomial of degree three, so and . This way, cannot vanish at every point of .
Choose in this grid with . By construction, no three points generated from are collinear. These points are also pairwise distinct: if for two distinct subsets, the determinant involving any third subset would be zero. Finally, since the singleton points are distinct, the are distinct as well. □
Remark 4.7. One can in fact take in Lemma 4.6 for some ; the point of the above proof is that such configurations are generic at lacunary scales.
Fix as in Lemma 4.6, and set
Then , , and no three points of are collinear. Moreover, the seed coordinates are integers of size , and the determinants associated with triples of distinct points of have absolute value at least .
For each and with , we have . Thus at least pairs satisfy . Let , where the supremum is over non-vertical affine lines. Define the representation function
and observe that
On the other hand, by Cauchy–Schwarz,
Therefore
In particular, the seed energy ratio can be made arbitrarily large by increasing .
Taking the -fold product, we define the product configuration to be the ordered pair
Its two components are subsets of and , respectively. In the rest of this paper, we identify with by grouping together the horizontal and vertical coordinates:
Now, fix some , in particular requiring (with from Lemma 4.3), and define the sets
Set
The parabolically rescaled and thickened product configuration is the ordered pair
Lemma 4.8. For , we have and for each , . Moreover, any line in intersects at most two rectangles in .
Proof. This all follows immediately from the choices of , and from letting be sufficiently large. Note that will be of size to guarantee the separation condition. □
Intersecting Product with Structure
Now, we perform Step (6).
Intersecting the two components of Product with and , respectively, gives the pair
For , let
be a non-vertical affine -line, and write . We define
Our goal is to prove the following energy estimate.
Lemma 4.9. We have
Proof. We first estimate the sizes of and .
Claim 4.10. The following bounds hold:
Proof. The set is a disjoint union of axis-parallel boxes indexed by . Each such box is of the form , so . By Lemma 4.3 we have
Since the boxes are pairwise disjoint,
and (4.8) follows. Similarly, is a disjoint union of product boxes indexed by , , and each has horizontal sides of length and vertical sides of length . The number of points in such a box is
Using the upper bound of Lemma 4.3 and summing over all boxes yields (4.9). □
We next estimate the number of points of that remain in after translation by , for each .
Claim 4.11. For every , we have
where and denotes coordinatewise squaring.
Proof. Fix , with . Since the intervals are disjoint, for each there is a unique such that
For each choice of subsets , set
Thus precisely when and . In particular,
We first check that these intersections are large. The seed identities and show that the centres of the two rectangles defining differ by
Since , choosing gives . Both rectangles have width and height , so has width at least and height at least .
We now count lattice points in the resulting product boxes. There are choices of for each , giving pairwise disjoint boxes . Indeed, and the rectangles are pairwise disjoint. Under (4.6), each product is , where have side lengths at least and , respectively. Applying Lemma 4.3 to both boxes, whose side lengths are at least for sufficiently large, gives
Every point counted above belongs to , since its coordinate pairs lie in . Moreover, (4.10) and give
Since , this shows that
The product boxes are indexed by the tuples with . Since they are disjoint, summing their contributions gives
where the last inequality follows from (4.9). □
We now prove the following energy lower bound.
Claim 4.12. .
Proof. Recall that
where is the representation function . Similarly to (4.3),
by Claim 4.11. By Cauchy–Schwarz, , and the claim follows. □
It remains to bound the number of points of on a non-vertical -line.
Claim 4.13. .
Proof. Fix a non-vertical -line
and consider its image under . In the th embedding, its points lie on the real line
By Lemma 4.8, this line intersects at most two of the rectangles . Thus the points of in the image of lie in at most product boxes . Under the identification (4.6), each such product box has the form , where has side lengths .
Write . Each determines at most one point of , since . Thus Lemma 4.3 gives
Summing over the at most product boxes and taking the supremum over all non-vertical -lines, we obtain
Finally, by Claims 4.10, 4.12 and 4.13 we conclude that
and the lemma follows.
Completing the proof
To complete the proof of Proposition 4.1, we need to project and parabolically rescale our sets and , and then exploit the separation properties of .
Choose some , and consider the sets
Note that for chosen sufficiently large, we have that and . Since is a field homomorphism, this also preserves both the cardinalities of the sets and their parabolic energy.
Lemma 4.14. There exists , depending on but not on , such that every difference between two elements of , vertical difference between two points of , or determinant formed by three points of is either zero or has absolute value at least .
Proof. By Lemma A.1, if , then
Recall that the seed coordinates are bounded by and was chosen sufficiently large in terms of . Thus whenever is a difference of algebraic coordinates of or , or a determinant formed from three algebraic pairs corresponding to points of .
For distinct with , observe that
where the last inequality follows from (21) applied to . We have analogous lower bounds for nonzero vertical differences in and determinants of -differences. Choosing completes the proof.
Let us now introduce the notation
In words, is the maximum number of points of that a strip with vertical width contains.
Lemma 4.15. We have .
Proof. Fix and an interval of length , and write
By the determinant conclusion of Lemma 4.14, every three points of are collinear. Moreover, any two distinct points of have different horizontal coordinates. Thus, if contains at least two points, they all lie on a single non-vertical real line .
Let be the algebraic pairs corresponding to two distinct points of . Since , we may define , . We can then lift to the -line
By collinearity and injectivity of , every point of lifts to a point of on . Therefore
The same inequality holds trivially if contains at most one point. Taking the supremum over completes the proof. □
Proof of Proposition 4.1. Fix as in Lemma 4.2. Next, fix some so large that the gain of Lemma 4.9 beats the by a definite factor; for example, choose so large that
where are as in Lemma 4.9, for every . Now, take as in (4.11). By Lemma 4.14, we see that is -separated, completing (i). By Lemma 4.15, condition (ii) holds with . Since the slicing and parabolic rescaling preserved cardinalities and energy, Lemma 4.9 alongside our choice of implies (iii). □
Proof of Theorem 1.3
By combining Propositions 3.1 and 4.1, we can complete the proof of Theorem 1.3.
Proof of Theorem 1.3. Let be as in Proposition 4.1. Let , and choose such that
where is fixed by Proposition 3.1. Now, set , and , where are defined as in Proposition 4.1. Then the sets produced by Proposition 4.1 can be plugged into Proposition 3.1 to obtain
But note that , where
which finishes the proof. □
For a hypersurface whose second fundamental form is nonzero at some point, explicit choices of parameters can be made in the previous proof to yield
for all sufficiently large . We made no attempt to optimise this exponent, in favour of expositional clarity; the exponent bound is clearly improvable by a more careful choice of parameters. On the other hand, it is unlikely that a similar approach could match the best known upper bound for every , proved by Carbery, Iliopoulou, and Wang [10] for strictly convex compact patches in ; indeed, there is no dependence on in (5.1).
We also note that the optimal power-loss exponent for a hypersurface , given by
depends on more than just the dimension ; indeed, for generic surfaces in one has , whereas [21 Theorem 4.2] implies that for the cone (which is rank-1), one has .
Appendix A. Number fields and lattices
In this appendix we collect definitions and standard identities used in Section 4; see, for instance, [24 Lectures 4, 6, 12 and 14].
Number fields and embeddings. A number field of degree is a field extension with . It has exactly distinct embeddings into , where an embedding is an injective field homomorphism fixing . The field is totally real if all these embeddings take values in . In this case, write them as . The Minkowski embedding is
Algebraic integers, norm and trace. An element of is an algebraic integer if it is a root of a monic polynomial with coefficients in . These elements form the ring of integers . As an additive group, has a -basis , called an integral basis, and is a full lattice in . The field norm and trace of are
Both belong to , and both are integers when . In particular, the norm of a nonzero algebraic integer is a nonzero integer.
Lemma A.1. For every nonzero , we have . Moreover, if for all , with , then for all .
Proof. Since the norm of a nonzero algebraic integer is a nonzero integer,
If for every , then, for each ,
Hence , as claimed. □
Discriminant and root discriminant. The discriminant of is
It is independent of the integral basis and is a positive integer because is totally real. The root discriminant is . The covolume of , namely the volume of a fundamental parallelepiped, satisfies
The dual lattice. For , the Euclidean dual lattice is
It satisfies . Indeed, if
then and hence . Moreover,
has integer entries, so has rational entries. Since , we obtain . Additionally, we have
Indeed, if we write for some invertible real matrix , the defining condition for is equivalent to . Thus , and consequently
Declaration of AI usage
An early version of the proof of a special case of Theorem 1.3 was generated by OpenAI’s ChatGPT-5 Sol and ChatGPT-6 Astra from the previous works of the current authors. Namely, in a series of conversations held between 17 August and 10 September 2026 (GPT-5 Sol chat, GPT-6 Astra chat 1, GPT-6 Astra chat 2, GPT-6 Astra chat 3) those models were asked to gradually improve the lower bounds on the local Mizohata–Takeuchi constants from the paper [3].3 The prompts also directed the models to [8, 9]. The original GPT construction relied on towers of number fields with prescribed prime splitting behaviour, provided by Hajir and Maire [13]. The authors then found a different, more direct proof presented here, which uses a different arithmetic construction in § 5 and generalises to hypersurfaces. ChatGPT-6 Astra was also used for copy-editing, for drafting the material in the short Appendix A on basic notions of algebraic number theory, and for proofreading of the final manuscript. The authors wrote the manuscript in their own words and assume full responsibility for its content.
References
(Hannah Cairo) Department of Mathematics, William E. Kirwan Hall, University of Maryland, College Park, MD, U.S. Email address: [email protected]
(Inbo Gottlieb Fenves) University of Wisconsin-Madison, Madison, WI, U.S. Email address: [email protected]
(Vjekoslav Kovač) University of Zagreb Faculty of Science, Department of Mathematics, Bijenička cesta 30, 10000 Zagreb, Croatia. Email address: [email protected]
(Itamar Oliveira) School of Mathematics, The Watson Building, University of Birmingham, Edgbaston, Birmingham, B15 2TT, England. Email address: [email protected], [email protected]
(Dmitrii Zakharov) Massachusetts Institute of Technology, Cambridge, MA, U.S. Email address: [email protected]
(Ruixiang Zhang) Department of Mathematics, Evans Hall, University of California, Berkeley, Berkeley CA, U.S. and School of Mathematics, 1 Einstein Drive, Institute for Advanced Study, Princeton NJ, U.S. Email address: [email protected], [email protected]
References
- [1]N. Alon, T. F. Bloom, W. T. Gowers, D. Litt, W. Sawin, A. Shankar, J. Tsimerman, V. Wang, and M. M. Wood, Remarks on the disproof of the unit distance conjecture, preprint (2026). arXiv:2605.20695.arxiv.org/abs/2605.20695
- [2]J. Bennett, S. Gutierrez, S. Nakamura, and I. Oliveira, A phase-space approach to weighted Fourier extension inequalities, Forum Math. Sigma 13 (2025), e181. doi:10.1017/fms.2025.10127. arXiv:2406.14886.DOI
- [3]J. Bennett, V. Kovač, S. Nakamura, and I. Oliveira, Rectangles, triangles and Schrödinger waves, preprint (2026). arXiv:2606.30178.
- [4]J. Bennett, S. Nakamura, Tomography bounds for the Fourier extension operator and applications, Math. Ann. 380 (2021), 119–159. doi:10.1007/s00208-020-02131-0.DOI
- [5]J. A. Barceló, A. Ruiz, and L. Vega, Weighted estimates for the Helmholtz equation and some applications, J. Funct. Anal. 150 (1997), no. 2, 356–382. doi:10.1006/jfan.1997.3131.DOI
- [6]T. Bloom, Sum-product, unit distances, and number fields, Erdős Problems, blog post, 31 May 2026. Online article.
- [7]T. F. Bloom, W. Sawin, C. Schildkraut, and D. Zhelelov, The sum-product conjecture is false for real numbers, preprint (2026). arXiv:2605.28781.arxiv.org/abs/2605.28781
- [8]H. Cairo, A counterexample to the Mizohata–Takeuchi conjecture, preprint (2025). arXiv:2502.06137.arxiv.org/abs/2502.06137
- [9]H. Cairo and R. Zhang, Power loss for the Mizohata–Takeuchi conjecture on Cᵏ convex hypersurfaces, preprint (2025). arXiv:2512.08064.arxiv.org/abs/2512.08064
- [10]A. Carbery, M. Iliopoulou, and H. Wang, Some sharp inequalities of Mizohata–Takeuchi-type, Rev. Mat. Iberoam. 40 (2024), no. 4, 1387–1418. doi:10.4171/RMI/1463. arXiv:2302.11877.DOI
- [11]I. G. Fenves, Cusp excursions, lattice points on manifolds, and the Mizohata–Takeuchi conjecture, preprint (2026). arXiv:2606.27020.arxiv.org/abs/2606.27020
- [12]L. Guth, An enemy scenario in restriction theory, joint talk for the AIM Research Community “Fourier restriction conjecture and related problems” and the HAPPY Network, 2 March 2022. Recorded lecture.
- [13]F. Hajir and C. Maire, On the invariant factors of class groups in towers of number fields, Canad. J. Math. 70 (2018), no. 1, 142–172. doi:10.4153/CJM-2017-032-9.DOI
- [14]J. Hao, R. Wang, and R. Zhang, Global Mizohata–Takeuchi for indicator weights of bounded complexity in R², preprint (2026). arXiv:2610.27659.arxiv.org/abs/2610.27659
- [15]J. Kim and L. Wang, A Mizohata–Takeuchi-type estimate for the hyperbolic paraboloid, preprint (2026). arXiv:2610.07909.arxiv.org/abs/2610.07909
- [16]J. Martinet, Tours de corps de classes et estimations de discriminants, Invent. Math. 44 (1978), no. 1, 65–73. doi:10.1007/BF01389902.DOI
- [17]P. Mattila, Fourier analysis and Hausdorff dimension. Cambridge Studies in Advanced Mathematics, 150, Cambridge, (2015).DOI
- [18]S. Mizohata, On some Schrödinger type equations, Proc. Japan Acad. Ser. A Math. Sci. 57 (1981), no. 2, 81–84. doi:10.3792/pjaa.57.81.DOI
- [19]OpenAI, Local smoothing for the Euclidean wave equation, research manuscript, 25 October 2026. Public manuscript.
- [20]OpenAI, The Kakeya maximal conjecture in three dimensions, research manuscript, 23 September 2026. Public manuscript.
- [21]A. Ortiz, A sharp weighted Fourier extension estimate for the cone in R³ based on circle tangencies, J. Anal. Math. 159 (2026), no. 2, 747–800. arXiv:2307.11731.arxiv.org/abs/2307.11731
- [22]D. Simmons and Y. Solomon, A Danzer Set for Axis Parallel Boxes, Proc. Amer. Math. Soc. 144 (2016), no. 6, 2725–2729. doi:10.1090/proc/12911.
- [23]E. M. Stein, Some problems in harmonic analysis, Proc. Sympos. Pure Math., Williamstown, Mass., (1978), 3–20.DOI
- [24]A. V. Sutherland, Number Theory I, MIT course 18.785, lecture notes, Fall 2021. Lecture notes.
- [25]J. Takeuchi, A necessary condition for H∞-wellposedness of the Cauchy problem for linear partial differential operators of Schrödinger type, J. Math. Kyoto Univ. 25 (1985), no. 3, 459–472.
- [26]R. Zhang, On Cᵏ convex variants of Stein’s weighted L² conjecture for Bochner–Riesz means, preprint (2026). arXiv:2602.027045.
- [27]R. Zhang, The Role of Some Geometric Objects in Fourier Restriction Theory, in Proceedings of the International Congress of Mathematicians 2026, vol. 4, Invited Lectures (Sections 5–8), 801–821. doi:10.1137/25M1806272.DOI