On L p density of self-similar measures and Bernoulli convolutions

arXiv:2611.17277 · math.DS, math.CA, math.PR · Submitted 2026-11-12 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Mathematics. The day in mathematics: new theorems, new proofs, and what they open.

Ada: I'm Ada.

Emmy: And I'm Emmy, here with Kolya.

Ada: We're looking at the paper: "On L p density of self-similar measures and Bernoulli convolutions".

Emmy: Here's the gist of the paper. On L p density of self similar measures and Bernoulli convolutions proves that an absolutely continuous self-similar measure has an Lp density for some p > 1, and establishes explicit examples of Bernoulli convolutions with a continuous density Fundamental Theorem on Density Theorem 1.1 states that if a probability measure ν is absolutely continuous with respect to the Lebesgue measure, then its density is in L p(R d) for some p > 1 The proof relies on Lemma 2.1, which is a version of the reverse Hölder criterion of Spadaro

Spa12: , as further developed by Kinnunen–Kurki–Mudarra

KKM22: This lemma assumes control over the mass of ν(E) for sets E of small measure and establishes a regularity gain for ν.

Ada: Explicit Examples of Bernoulli Convolutions The paper applies quantitative proof methods to establish explicit examples of Bernoulli convolutions with a continuous density going beyond only the previously known case of Garsia numbers Theorem 1.2 provides a condition for an algebraic number λ to ensure absolute continuity with a density in Cc(R).

Emmy: Specifically, if λ ∈ (0, 1) is algebraic of degree dλ, has Mahler measure Mλ and satisfies λ ≥ 1 − min

1, log Mλ: 1 + dλ + log Mλ100, then νλ is absolutely continuous with a density in Cc(R).

Ada: Quantitative Bounds for Bernoulli Convolutions The paper introduces Theorem 1.3, which provides an explicit non-Garsia example of a Bernoulli convolution with a continuous density.

Paper summary: Ada: So we're looking at this paper called "On L p density of self-similar measures and Bernoulli convolutions". Basically, the main idea here is proving that if you have a self-similar measure that's absolutely continuous with respect to the Lebesgue measure, then it has an L p density for some exponent p greater than one.

Emmy: And they go further by establishing explicit examples of Bernoulli convolutions where you can actually find a continuous density. That's pretty significant because those convolutions can be really tricky to handle analytically.

Kolya: So, what does this mean for the actual structure of these measures? Is it just that they have *some* density, or is there a specific kind of regularity we're getting here?

Ada: It’s more than just some density; the theorem shows that this L p density is linked to how much mass the measure puts on small sets. They use a lemma based on what they call the reverse Hölder criterion from Spadaro, which helps them get a regularity gain for this measure (Fal14,BSS23).

Emmy: And then they take that general result and apply it quantitatively to construct actual examples of Bernoulli convolutions with continuous densities going beyond just the known cases related to Garsia numbers. That’s what they call Theorem one point two <ref:2611.17277#pg2>.

Kolya: So, the explicit examples are key here because theoretical existence is one thing, but showing you can actually write down a density function makes it concrete for application. What kind of conditions are they setting up for those examples?

Ada: They set up a condition on an algebraic number lambda that determines if the Bernoulli convolution with that parameter has a continuous density in Cc(R). Specifically, if lambda is algebraic of degree d lambda and has Mahler measure M lambda, and it satisfies this inequality: one minus the minimum of one and the logarithm of M lambda raised to the power one plus d lambda plus the logarithm of M lambda to the power one hundred then their measure has a continuous density in Cc(R) <ref:2611.17277#pg3>.

Emmy: That inequality looks quite specific. It shows how algebraic properties like degree and Mahler measure combine with lambda to guarantee that nice density. It’s a structural condition they are using to pin down the required smoothness of the resulting distribution.

Paper summary: Kolya: If we look at Theorem one point three, it seems like they found a way to get an L two density for nu lambda by meeting a slightly different bound on lambda <ref:2611.17277#pg2>. That means the measure squared has an L two density, which implies the original measure has a continuous density in Cc(R) <ref:2611.17277#pg2>.

Ada: Exactly. Theorem one point three states that if lambda is algebraic of degree d lambda, Mahler measure M lambda, and satisfies lambda greater than or equal to one minus the minimum of one and the logarithm of M lambda raised to the power one plus d lambda plus the logarithm of M lambda cubed, then the square of their Bernoulli convolution has an L two density in R <ref:2611.17277#pg2>.

Emmy: That cubic exponent on Mahler measure is what they used to establish that specific result. It suggests that for certain algebraic numbers, this relationship between lambda and its size dictates the regularity we see in the resulting density.

Kolya: And when they look at rational numbers, Corollary one point four gives a simpler condition using p and q for lambda equal to one minus p over q <ref:2611.17277#pg2>. If p is less than or equal to q squared plus the logarithm of q cubed, then their Bernoulli convolution has a continuous density in Cc(R).

Ada: So we have these different conditions—one involving the power one hundred and another involving a power three—which are derived from different quantitative arguments <ref:2611.17277#pg3>. It shows that the explicit construction of these examples requires precise control over how the measure behaves at small scales.

Emmy: The paper mentions using Young’s inequality for scaled convolutions in Lemma three point two and arithmetic input estimates in Lemma three point three to get those bounds on B lambda, which is related to how much mass is concentrated <ref:2611.17277#pg3>. That’s where the algebraic geometry of the underlying structure really comes into play with these probabilistic tools.

Kolya: And then there's this section about effective detail estimation, Proposition three point eight, which shows that if you meet certain detail bounds for a probability measure sigma, it forces it to have a density f such that the difference between f and the convolution of sigma with some function eta is very small in an L one sense.

Ada: That smallness bound is what lets them conclude that their measure has an L one density at those scales, which leads directly to the result in Proposition three point one that requires it to be in L one plus delta lambda of R <ref:2611.17277#pg3>. It’s a chain of reasoning connecting scale control to density existence.

Paper summary: Emmy: The entropy and variance bounds they use, like Lemma three point nine and Lemma three point eight, are crucial for controlling the behavior across those different scales effectively enough so that the final convolution result holds true. It’s about making sure the error term doesn't blow up as you look at finer details of the fractal construction.

Kolya: So, what’s a real limitation they point out? They mention that Theorem one point two relies on controlling the measure from a fixed scale onwards, which requires explicit separation bounds established in Lemma three point one two to compare against previous estimates <ref:2611.17277#pg3,the measure from a fixed scale onwards>. That’s where they have to be very careful with how they set up the comparison between different scales.

Ada: And that control over separation bounds is what makes Theorem one point two depend on the degree of lambda, because you need that explicit separation information for Lemma two point one to work as intended <ref:2611.17277#pg2>. It shows that the required conditions aren't just about size; they are tied to the algebraic complexity of lambda.

Emmy: The paper uses Lean formalization for all these results, which is a big part of its contribution, showing how rigorous they want this relationship between algebra and analysis to be. It’s a very thorough approach to establishing these explicit examples.

Kolya: For someone who just wants the practical implication, it means we can now precisely identify *which* algebraic numbers will yield Bernoulli convolutions with continuous densities, rather than just knowing that some might exist. That shifts the problem from existence to construction under specific number-theoretic constraints.

Ada: So, in conclusion for this paper "On L p density of self-similar measures and Bernoulli convolutions", they prove the fundamental property that absolutely continuous self-similar measures have an L p density for some p greater than one, and they provide explicit conditions—like the ones involving Mahler measure and degree—that guarantee Bernoulli convolutions with a continuous density.

Emmy: It’s about connecting the deep structure of algebraic numbers to the regularity of geometric measures. The authors use quantitative methods to establish these explicit examples, which is a solid way to move past purely theoretical existence proofs.

Kolya: It suggests that the question isn't just whether these measures are continuous or singular, but how much regularity we can expect based on the arithmetic properties of the similarity transformation parameters.

Ada: So they’ve given us concrete, explicit conditions for when you can have a continuous density in Cc(R) for Bernoulli convolutions. That’s what this paper delivers in terms of new results and tools.

Conclusion: Ada: So we're looking at "On L p density of self-similar measures and Bernoulli convolutions," and this paper is about proving that if you have a self-similar measure that's absolutely continuous with respect to the Lebesgue measure, it must have an L p density for some p greater than one.

Emmy: And they’re providing specific examples of those Bernoulli convolutions where we can actually find a continuous density function.

Kolya: What does that mean for the actual structure of these measures? Is it just that they have *some* density, or is there a specific kind of regularity we're getting here?

Ada: It’s more than just some density; the theorem shows that this L p density is linked to how much mass the measure puts on small sets. They use a lemma based on what they call the reverse Hölder criterion from Spadaro, which helps them get a regularity gain for this measure.

Emmy: And then they take that general result and apply it quantitatively to construct actual examples of Bernoulli convolutions with continuous densities going beyond just the known cases related to Garsia numbers. That’s what Theorem one point two does.

Kolya: So, the explicit examples are key here because theoretical existence is one thing, but showing you can actually write down a density function makes it concrete for application. What kind of conditions are they setting up for those examples?

Ada: They set up a condition on an algebraic number lambda that determines if the Bernoulli convolution with that parameter has a continuous density in Cc(R). Specifically, if lambda is algebraic of degree d lambda and has Mahler measure M lambda, and it satisfies this inequality involving one minus the minimum of one and the logarithm of M lambda raised to the power one plus d lambda plus the logarithm of M lambda to the power one hundred then their measure has a continuous density in Cc(R).

Emmy: That inequality looks quite specific. It shows how algebraic properties like degree and Mahler measure combine with lambda to guarantee that nice density. It’s a structural condition they are using to pin down the required smoothness of the resulting distribution.

Kolya: If we look at Theorem one point three, it seems like they found a way to get an L two density for nu lambda by meeting a slightly different bound on lambda. That means the measure squared has an L two density, which implies the original measure has a continuous density in Cc(R).

Wrap-up: Ada: Exactly. Theorem one point three states that if lambda is algebraic of degree d lambda, Mahler measure M lambda, and satisfies lambda greater than or equal to one minus the minimum of one and the logarithm of M lambda raised to the power one plus d lambda plus the logarithm of M lambda cubed, then the square of their Bernoulli convolution has an L two density in R.

Emmy: That cubic exponent on Mahler measure is what they used to establish that specific result. It suggests that for certain algebraic numbers, this relationship between lambda and its size dictates the regularity we see in the resulting density.

Kolya: And when they look at rational numbers, Corollary one point four gives a simpler condition using p and q for lambda equal to one minus p over q. If p is less than or equal to q squared plus the logarithm of q cubed, then their Bernoulli convolution has a continuous density in Cc(R).

Ada: So we have these different conditions—one involving the power one hundred and another involving a power three—which are derived from different quantitative arguments. It shows that the explicit construction of these examples requires precise control over how the measure behaves at small scales.

Emmy: The paper mentions using Young’s inequality for scaled convolutions in Lemma three point two and arithmetic input estimates in Lemma three point three to get those bounds on B lambda, which is related to how much mass is concentrated. That’s where the algebraic geometry of the underlying structure really comes into play with these probabilistic tools.

Kolya: And then there's this section about effective detail estimation, Proposition three point eight, which shows that if you meet certain detail bounds for a probability measure sigma, it forces it to have a density f such that the difference between f and the convolution of sigma with some function eta is very small in an L one sense.

Ada: That smallness bound is what lets them conclude that their measure has an L one density at those scales, which leads directly to the result in Proposition three point one that requires it to be in L one plus delta lambda of R. It’s a chain of reasoning connecting scale control to density existence.

Wrap-up: Emmy: The entropy and variance bounds they use, like Lemma three point nine and Lemma three point eight, are crucial for controlling the behavior across those different scales effectively enough so that the final convolution result holds true. It’s about making sure the error term doesn't blow up as you look at finer details of the fractal construction.

Kolya: So what’s a real limitation they point out? They mention that Theorem one point two relies on controlling the measure from a fixed scale onwards, which requires explicit separation bounds established in Lemma three point one two to compare against previous estimates. That’s where they have to be very careful with how they set up the comparison between different scales.

Ada: And that control over separation bounds is what makes Theorem one point two depend on the degree of lambda, because you need that explicit separation information for Lemma two point one to work as intended. It shows that the required conditions aren't just about size; they are tied to the algebraic complexity of lambda.

Emmy: The paper uses Lean formalization for all these results, which is a big part of its contribution, showing how rigorous they want this relationship between algebra and analysis to be. It’s a very thorough approach to establishing these explicit examples.

Kolya: For someone who just wants the practical implication, it means we can now precisely identify *which* algebraic numbers will yield Bernoulli convolutions with continuous densities, rather than just knowing that some might exist. That shifts the problem from existence to construction under specific number-theoretic constraints.

Ada: So in conclusion for this paper "On L p density of self-similar measures and Bernoulli convolutions," they prove the fundamental property that absolutely continuous self-similar measures have an L p density for some p greater than one, and they provide explicit conditions—like the ones involving Mahler measure and degree—that guarantee Bernoulli convolutions with a continuous density.

Emmy: It’s about connecting the deep structure of algebraic numbers to the regularity of geometric measures. The authors use quantitative methods to establish these explicit examples, which is a solid way to move past purely theoretical existence proofs.

Kolya: It suggests that the question isn't just whether these measures are continuous or singular, but how much regularity we can expect based on the arithmetic properties of the similarity transformation parameters.

Ada: So they’ve given us concrete, explicit conditions for when you can have a continuous density in Cc(R) for Bernoulli convolutions. That’s what this paper delivers in terms of new results and tools. Now, we look at how these algebraic constraints affect our understanding of fractal geometry and measure theory.

Samuel Kittle, Constantin Kogler

University College London · Institute for Advanced Study

math.DS, math.CA, math.PR

Submitted: 2026-11-12

Updated: 2026-11-12

Comments: 18 pages

Code: https://github.com/ckkogler/l1-to-lp-supplementary-material

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 81/100

The gist: On L p density of self similar measures and Bernoulli convolutions proves that an absolutely continuous self-similar measure has an Lp density for some p > 1…

Key concepts

Lp Density
This refers to the existence of a function (density) that describes how the probability mass is distributed over space, such that the integral of this density raised to the power p is finite. The theorem establishes that for certain self-similar measures, this density exists even when p is greater than 1.
Bernoulli Convolution
This measure is created by repeatedly convolving an initial probability measure with itself, scaled by a parameter. The paper focuses on finding specific algebraic numbers (like $\lambda$) that control the properties of this convolution to ensure it has a continuous density.
Absolute Continuity
A measure is absolutely continuous with respect to the Lebesgue measure if it can be described by a function (its density) such that any set with zero Lebesgue measure has zero probability mass under the measure. The paper seeks conditions on $\lambda$ that guarantee this property for Bernoulli convolutions.
Reverse Hölder Criterion
This is a mathematical tool used in the proof, which provides a regularity gain for measures. It helps establish control over the mass of small sets within the measure, which is essential for proving that the measure possesses sufficient smoothness to have an Lp density.

Terminology

Summary

On L p density of self similar measures and Bernoulli convolutions proves that an absolutely continuous self-similar measure has an Lp density for some p > 1, and establishes explicit examples of Bernoulli convolutions with a continuous density

Fundamental Theorem on Density

Theorem 1.1 states that if a probability measure ν is absolutely continuous with respect to the Lebesgue measure, then its density is in L p(R d) for some p > 1 The proof relies on Lemma 2. p. 0

Where the paper says it

Let µ be a probability measure supported on finitely many contracting similarities of R d for d ≥ 1 and denote by ν the self-similar measure of µ. If ν is absolutely continuous with respect to the Lebesgue measure, then the density of ν is in L p (R d ) for some p > 1.

1, which is a version of the reverse Hölder criterion of Spadaro [Spa12], as further developed by Kinnunen–Kurki–Mudarra [KKM22] This lemma assumes control over the mass of ν(E) for sets E of small measure and establishes a regularity gain for ν. p. 0
Where the paper says it

1, which is a version of the reverse H¨older criterion of Spadaro [Spa12], as further developed by Kinnunen–Kurki–Mudarra [KKM22]. 1 assumes for an absolutely continuous measure ν on R d control over the mass of ν(E) for sets E of small measure and then establishes a regularity gain for ν.

Explicit Examples of Bernoulli Convolutions

The paper applies quantitative proof methods to establish explicit examples of Bernoulli convolutions with a continuous density going beyond only the previously known case of Garsia numbers Theorem 1.2 provides a condition for an algebraic number λ to ensure absolute continuity with a density in Cc(R). p. 1

Where the paper says it

Garsia [Gar62] proved in 1962 that νλ has an L∞ density if λ ∈ (1/2, 1) is a Garsia number, that is, λ satisfies Mλ = 2. There was no progress on understanding absolute continuity for explicit λ until Varj´u [Var19] proved in 2019 that νλ is absolutely continuous if λ > 1 − c min{log Mλ,(log Mλ) −2} for c > 0 an absolute constant.

Specifically, if λ ∈ (0, 1) is algebraic of degree dλ, has Mahler measure Mλ and satisfies λ ≥ 1 − min[1, log Mλ]1 + dλ + log Mλ100, then νλ is absolutely continuous with a density in Cc(R). p. 1
Where the paper says it

If λ ∈ (1/2, 1) is algebraic of degree dλ, has Mahler measure Mλ and satisfies λ ≥ 1 − min{1, log Mλ} 1 + dλ + log Mλ 100 , (1. 2) then νλ is absolutely continuous with a density in Cc(R).

Quantitative Bounds for Bernoulli Convolutions

The paper introduces Theorem 1.3, which provides an explicit non-Garsia example of a Bernoulli convolution with a continuous density. p. 1

Where the paper says it

Yu [Yu22] proved in 2022 that νλ has a continuous density if λ is a Garsia number satisfying λ ∈ [1/ √ 2, 1). In this paper, we give the first explicit non-Garsia examples of Bernoulli convolutions with a continuous density.

This theorem states that if λ satisfies λ ≥ 1 − min[1, log Mλ]1 + dλ + log Mλ3, then νλ2 is absolutely continuous with a density in L 2(R) and therefore νλ has a density in Cc(R). p. 2
Where the paper says it

([KKa]) If λ ∈ (1/2, 1) is algebraic of degree dλ, has Mahler measure Mλ and satisfies λ ≥ 1 − min{1, log Mλ} 1 + dλ + log Mλ 3 , (1. 3) then νλ2 is absolutely continuous with a density in L 2 (R) and therefore νλ has a density in Cc(R).

Corollary 1. p. 2
Where the paper says it

3 to rational numbers, one obtains the following. Corollary 1.

4 shows that for rational numbers λ = 1 − p/q ∈ (0, 1), if p ≤ q2 + log q3, then νλ is absolutely continuous with a density in Cc(R). p. 0
Where the paper says it

In this paper, we establish the fundamental property that an absolutely continuous self-similar measure has a density in L p for some p > 1. Moreover, we apply the quantitative proof method of the latter result to establish explicit examples of Bernoulli convolutions with a continuous density going beyond the only previously known case of Garsia numbers.

Technical Tools and Proof Structure

The proof of Theorem 1.2 utilizes Young’s inequality for scaled convolutions, as detailed in Lemma 3. p. 5

Where the paper says it

We will use Young’s inequality in the following form. 2 (Young’s inequality for scaled convolutions).

2. This is combined with arithmetic input estimates from Lemma 3. p. 5
Where the paper says it

□ The following basic estimate on Bλ is used as well. 3 (Arithmetic input).

3, which establishes that Bλ ≥ 3 and log1/(1 − λ) ≤ Bλ. p. 4
Where the paper says it

Bernoulli convolutions: Proof of Theorem 1. 2 Throughout this section, write dλ for the degree of an algebraic number λ ∈ (0, 1) and χλ = − log λ and Bλ = 1 + dλ + log Mλ min{1, log Mλ} .

Furthermore, the proof of Proposition 3. p. 5
Where the paper says it

2 and Section 3. The proof of Proposition 3.

1 relies on controlling measure at fixed explicit scales using Lemma 3. p. 2
Where the paper says it

1 one needs to control the measure from a fixed scale onwards. This requires explicit separation bounds, which we establish in Lemma 3.

4. This involves establishing effective detail bounds through Lemma 3. p. 7
Where the paper says it

□ Next we recall the definitions of detail and higher order detail from [Kit24,Kit25]. We will use bounds on detail to establish the assumptions of Lemma 3.

6 and subsequent propositions like Proposition 3. p. 4
Where the paper says it

All logarithms are natural. The main result of this section is the following proposition.

7.

Effective Detail Estimation

Proposition 3.8 shows that for a probability measure σ, if certain detail bounds are met, then σ is absolutely continuous with density f such that f − σ ∗ η11 ≤ 4δR0 + 1/4R0. p. 8

Where the paper says it

Suppose that a probability measure σ satisfies for R ≥ R0, 1 ≤ k ≤ 10 log R and r ∈ (0, 1) that s (k) e−R (σ) ≤ 100−k and sr(σ) ≤ δ. 8) Then σ is absolutely continuous with density f and ∥f − σ ∗ η1∥1 ≤ 4δR0 + 1 4R0 .

This is derived by applying Lemma 3. p. 4
Where the paper says it

□ Theorem 1. 1 follows by applying Lemma 2.

5 and Lemma 3. p. 0
Where the paper says it

1 is short and combines self-similarity with a general higher integrability result. The key lemma of the proof of Theorem 1.

6 to obtain the required assumptions for Lemma 3. p. 7
Where the paper says it

□ Next we recall the definitions of detail and higher order detail from [Kit24,Kit25]. We will use bounds on detail to establish the assumptions of Lemma 3.

4. This process ultimately leads to the conclusion that fλ ∈ L1 + δλ(R) as required by Proposition 3. p. 9
Where the paper says it

5) and Lemma 3. 4 gives fλ ∈ L 1+δλ (R), as required.

1. The paper also details the use of entropy and variance to establish necessary bounds, such as Lemma 3. p. 7
Where the paper says it

□ Next we recall the definitions of detail and higher order detail from [Kit24,Kit25]. We will use bounds on detail to establish the assumptions of Lemma 3.

9 and Lemma 3. p. 0
Where the paper says it

1 is short and combines self-similarity with a general higher integrability result. The key lemma of the proof of Theorem 1.

8.

Entropy Gap for Variance Control

Lemma 3.13 provides a bound on the variance in a block, stating that every integer l ≥ 100B2λ satisfies Z e−l/(32Bλ) e−2B2λl V (Xl; s) ds/s ≥ l1010B7λ. p. 9

Where the paper says it

12 and compare to previous separation estimates right after stating Lemma 3. We first bound detail in terms of conditional variance as stated in Lemma 3.

This inequality is derived by estimating the difference in entropy over a range of scales, as shown in Lemma 3. p. 4
Where the paper says it

Write qn = γ1 · · · γn and consider the stopping time τr = min{n ≥ 1 : ρ(qn) ≤ r} and note ρ∗r < ρ(qτr ) ≤ r. It is well known that qτr ν has the same distribution as ν, as is shown in [Kit25, Lemma 2.

11. This bound is crucial for bounding the order k detail of the sum of blocks Yl,v by 100−k. p. 9
Where the paper says it

We first bound detail in terms of conditional variance as stated in Lemma 3. Then, we use entropy and algebraic separation to find enough conditional variance in disjoint blocks of the random power series determining the Bernoulli convolution.

The final step confirms that convolution with the sum of these blocks proves (3. p. 9
Where the paper says it

We first bound detail in terms of conditional variance as stated in Lemma 3. Then, we use entropy and algebraic separation to find enough conditional variance in disjoint blocks of the random power series determining the Bernoulli convolution.

11).

Machine Contribution Statement

The main idea of this paper is due to GPT6 Astra, which asked for explicit examples of self-similar measures with an L p density resulting in the discovery of Theorem 1.1. Various simplifications led to Lemma 2. p. 2

Where the paper says it

) asked GPT-6 Astra to find explicit examples of self-similar measures with an L p density resulting in the discovery of Theorem 1. Various simplifications led to Lemma 2.

1. Realising that Theorem 1. p. 2
Where the paper says it

Various simplifications led to Lemma 2. Realising that Theorem 1.

1 is quantitative, numerous conversations with GPT-6 Astra led to Theorem 1. p. 2
Where the paper says it

Realising that Theorem 1. 1 is quantitative, numerous conversations with GPT-6 Astra led to Theorem 1.

2. The paper is written by the authors and we take responsibility for its correctness. p. 2
Where the paper says it

1 is quantitative, numerous conversations with GPT-6 Astra led to Theorem 1. The paper is written by the authors and we take responsibility for its correctness.

Lean Formalization

GPT-6 Astra formalized all results from this paper and all necessary results from previous work, including Theorem 1.3. The Lean source code is available in the accompanying repository. p. 2

Where the paper says it

GPT-6 Astra formalized all results from this paper and all necessary results from previous work, including Theorem 1. The Lean source code is available in the accompanying repository [KKb].

Acknowledgments

The first-named author (S.K.) gratefully acknowledges support from the Heilbronn Institute for Mathematical Research. p. 2

Where the paper says it

The first-named author (S. ) gratefully acknowledges support from the Heilbronn Institute for Mathematical Research.

The second-named author (C. p. 2
Where the paper says it

The main idea of this paper is due to GPT6 Astra. On September 10th 2026, the second-named author (C.

p. 2
Where the paper says it

The main idea of this paper is due to GPT6 Astra. On September 10th 2026, the second-named author (C.

K.) holds a Postdoc Mobility Fellowship from the Swiss National Science Foundation (grant number 235409) and thanks the Institute for Advanced Study. p. 2
Where the paper says it

The second-named author (C. ) holds a Postdoc Mobility Fellowship from the Swiss National Science Foundation (grant number 235409) and thanks the Institute for Advanced Study.

The second-named author (C.K.) thanks Elon Lindenstrauss for his comments and interest in this work. p. 2
Where the paper says it

The second-named author (C. ) thanks Elon Lindenstrauss for his comments and interest in this work.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Improved density estimation for self-similar measures by applying Lemma 2.1 to establish an Lp density for some p greater than 1 for any absolutely continuous self-similar measure. This allows explicit construction of Bernoulli convolutions with continuous densities beyond previously known cases like Garsia numbers, as shown in Theorem 1.2 and Theorem 1.3.

  2. Quantitative bounds on the exponent p are derived from the mass that a self-similar measure assigns to sets of small Lebesgue measure, enabling the use of recent methods for establishing explicit examples of absolutely continuous self-similar measures to obtain suitable measure estimates yielding quantitative Lp results.

  3. The system can now determine if a given algebraic parameter lambda satisfies Theorem 1.2 by checking the condition lambda greater than or equal to 1 minus min one log M lambda one plus d lambda plus log M lambda over 100, where d and M are derived from the polynomial defining the algebraic number.

  4. The system can verify if a Bernoulli convolution with parameter lambda has a continuous density by checking if its Mahler measure satisfies the specific inequality in Theorem 1.2, which involves parameters like degree d lambda and Mahler measure M lambda.

  5. The AI can utilize Lemma 3.2 to determine the density of the convolution of scaled measures (Daj ν) when the original measure has a density in L p for some p greater than or equal to k over k minus 1, ensuring the resulting function is uniformly continuous and compactly supported.

  6. The system can implement Lemma 3.4 to refine an existing density function by finding a function g lambda satisfying the condition ||f lambda - g lambda|| one divided by two times two one two and ||g lambda|| infinity less than or equal to one over two, which guarantees the resulting density f lambda is in L 1 plus delta lambda R.

  7. The system can estimate the effective detail of a measure using Lemma 3.6 to prove that a probability measure sigma is absolutely continuous with density f if its detail bounds satisfy s(k) e minus R sigma less than or equal to one hundred minus k for sufficiently large R.

  8. The AI can calculate the conditional variance of sums of random variables, as shown in Lemma 3.10, to bound the order k detail of the sum of blocks Y l v by one hundred minus k whenever specific conditions on conditional variances are met and a greedy algorithm is employed.

Abstract

We prove that an absolutely continuous self-similar measure has an L p density for some p>1. Moreover, explicit examples of Bernoulli convolutions with a continuous density are established.

Sources

Related papers