Elementary Bounds on Digital Sums of Powers, Factorials, and LCMs

Basic Information

• Paper ID: 2511.15850
• Title: Elementary Bounds on Digital Sums of Powers, Factorials, and LCMs
• Author: David G. Radcliffe
• Classification: math.NT (Number Theory)
• Publication Date: November 19, 2025
• Paper Link: https://arxiv.org/abs/2511.15850

Abstract

This paper establishes that the $b$-adic digit sum of $a^n$ grows at least logarithmically when $\log(d)/\log(b)$ is irrational, where $d$ is the smallest divisor of $a$ satisfying $\gcd(a/d, b) = 1$. The research methodology employs only elementary number theory and applies to a broad class of sequences, including factorials and $\Lambda(n) = \text{lcm}(1, 2, \ldots, n)$. The paper concludes with an illustrative proof of a known result: the $b$-adic digit sum of $a^n$ tends to infinity if and only if $\log(a)/\log(b)$ is irrational.

Research Background and Motivation

Core Problem

The core problem studied in this paper originates from a question posed by Polish mathematician Sierpiński in 1970: prove that the decimal digit sum of $2^n$ grows without bound as $n$ increases. While this problem appears simple, it carries profound significance in number theory.

Problem Significance

1. Non-monotonicity Challenge: Although $2^n$ grows rapidly, its digit sum sequence is not monotonically increasing (e.g., $2^4=16$ has digit sum 7, while $2^5=32$ has digit sum 5). Thus, proving unboundedness alone is insufficient to establish divergence to infinity.
2. Universality: This problem extends beyond $2^n$ to the general form $a^n$ in arbitrary base $b$, possessing broad theoretical significance.
3. Digit Distribution Theory: While the conjecture that the decimal digit sum of $2^n$ is approximately $4.5n\log_{10}2$ (based on uniform digit distribution) remains unproven, this stronger conjecture has not yet been established.

Limitations of Existing Research

• Senge-Straus (1973): Proved that $c_b(a^n) \to \infty$ if and only if $\log(a)/\log(b)$ is irrational, but provided no lower bound on growth rate.
• Stewart (1980): Established the lower bound $c_b(a^n) > \frac{\log n}{\log\log n + C} - 1$ under general conditions.
• Sanna (2015): Provided stronger bounds for factorials and LCM: $s_b(n!) > C\log n\log\log\log n$.

Novelty of This Work

This paper employs purely elementary number-theoretic methods (avoiding transcendental number theory and other advanced tools) to obtain logarithmic lower bounds $c_b(a^n) > C\log n$ under specific conditions, with methods generalizable to factorials, LCMs, and other sequences.

Core Contributions

1. Establishment of Logarithmic Lower Bounds: Under the condition that $\log(d)/\log(b)$ is irrational, proves $c_b(a^n) > C\log n$ (Theorem 4).
2. Systematization of Elementary Methods: Develops elementary proof techniques based on divisibility properties, avoiding transcendental number-theoretic tools like Baker's theorem (in the first four sections).
3. Broad Applicability: Extends the method to:
   • Factorial sequences: $c_b(n!) > C\log n$ (Theorem 5)
   • LCM sequences: $c_b(\Lambda_n) > C\log\log n$ (Theorem 6)
4. Complete Theoretical Picture: Section 5 employs Baker's theorem to provide an illustrative proof for the general case, reproducing results of Senge-Straus and Stewart.
5. Pedagogical Value: Beginning with the Sierpiński problem and progressively generalizing, the paper provides clear intuition and multiple exercises, demonstrating excellent pedagogical merit.

Methodology Details

Task Definition

Notation:

• $s_b(n)$: sum of digits of $n$ in base $b$
• $c_b(n)$: count of nonzero digits in the base-$b$ representation of $n$
• $\nu_p(n)$: exponent of prime $p$ in the prime factorization of $n$
• Since $c_b(n) \leq s_b(n) \leq (b-1)c_b(n)$, the two are asymptotically equivalent; thus the focus is on $c_b(n)$

Core Task: For a given sequence of positive integers $(a_n)$, determine the lower bound on the growth rate of $c_b(a_n)$.

Core Technical Framework

1. Basic Idea (Section 2: Decimal Case for $2^n$)

Key Observation: A positive multiple of a positive integer cannot be smaller than the integer itself.

Construction Method:

• Write the decimal representation of $2^n$ as $2^n = \sum_{i=0}^{\infty} d_i 10^i$
• Examine $2^n \bmod 10^{e(k)}$ (the last $e(k)$ digits)
• If $2^n$ is divisible by $2^{e(k)}$, then the number formed by these $e(k)$ digits is also divisible by $2^{e(k)}$
• By induction, partition digits into non-overlapping blocks, each containing at least one nonzero digit

Theorem 1 (Formalized): Let the sequence $(e(k))_{k\geq 1}$ satisfy $e(1) \geq 1$ and $2^{e(k)} > 10^{e(k-1)}$. If $n$ is divisible by $2^{e(k)}$ but not by 10, then $c_{10}(n) \geq k$.

Corollary 1: For positive integers $a$ divisible by 2 but not by 10, $c_{10}(a^n) \geq \log_4(n)$.

Proof Technique: Choose $e(k) = 4^{k-1}$, so $2^{e(k)} = 2^{4^{k-1}} > 10^{4^{k-2}} = 10^{e(k-1)}$ (for $k \geq 2$).

2. Generalization to Arbitrary Bases (Section 3)

Theorem 2 (General Base Version): Let $b \geq 2$ not be a prime power, and let $p$ be a prime divisor of $b$. If $\nu_p(n) \geq e(k)$ and $b \nmid n$, then $c_b(n) \geq k$.

Key Innovation—Correction Function $\xi$: To handle trailing zeros (i.e., the case $b | n$), introduce the function: $\xi(n) = \nu_p(n) - \nu_q(n) \cdot \frac{\nu_p(b)}{\nu_q(b)}$

where $p, q$ are distinct prime divisors of $b$. This function satisfies $\xi(b^r u) = \xi(u)$, i.e., is insensitive to trailing zeros.

Theorem 3 (Improved Version): If $\xi(n) \geq e(k)$, then $c_b(n) \geq k$. In particular, if $\xi(a_n) \to \infty$, then $c_b(a_n) \to \infty$.

3. Main Result (Theorem 4)

Theorem 4: Let $a \geq 2, b \geq 2$. Let $d$ be the smallest divisor of $a$ such that $\gcd(a/d, b) = 1$. If $\log(d)/\log(b)$ is irrational, then: $c_b(a^n) > C\log n$ where $C > 0$ depends only on $a$ and $b$.

Proof Strategy:

1. Decompose $b$ and $d$ into prime factors: $b = p_1^{e_1} \cdots p_t^{e_t}$, $d = p_1^{f_1} \cdots p_t^{f_t}$
2. If $\log(d)/\log(b)$ is irrational, then the ratios $f_i/e_i$ are not all equal
3. There exist primes $p = p_i, q = p_j$ such that $f_i/e_i > f_j/e_j$, thus $\xi(a) > 0$
4. Choose $r = \lceil \log_p b \rceil$, $e(k) = r^{k-1}$
5. For given $n$, take $k = \lceil \log_r \xi(a^n) \rceil = \lceil \log_r(n\xi(a)) \rceil$
6. By Theorem 3, $c_b(a^n) \geq k = \Theta(\log n)$

Generalization Applications

Factorial Sequences (Theorem 5)

Theorem 5: If $b$ has prime divisors $p, q$ satisfying $(p-1)\nu_p(b) \neq (q-1)\nu_q(b)$, then: $c_b(n!) > C\log n$

Key to Proof:

• Use Legendre's formula: $\nu_p(n!) = \frac{n - s_p(n)}{p-1}$
• Compute $\xi(n!) = n\left(\frac{1}{p-1} - \frac{\nu_p(b)}{(q-1)\nu_q(b)} + o(1)\right) = \Theta(n)$
• Apply Theorem 3

LCM Sequences (Theorem 6)

Theorem 6: If $b \geq 2$ is not a prime power, then: $c_b(\Lambda_n) > C\log\log n$

Key to Proof:

• Use $\nu_p(\Lambda_n) = \lfloor \log_p(n) \rfloor$
• Compute $\xi(\Lambda_n) = \Theta(\log n)$
• Apply Theorem 3 to obtain $c_b(\Lambda_n) = \Theta(\log\log n)$

Complete Proof for General Case (Section 5)

Using Baker's theorem (a tool from transcendental number theory), the most general result is proved:

Theorem 8: If $\log(a)/\log(b)$ is irrational, then for sufficiently large $n$: $c_b(a^n) > \frac{\log n}{\log\log n + C}$

Proof Strategy:

1. Write the base-$b$ representation of $a^n$ as blocks of digits
2. Estimate the ratio $m(i+1)/m(i)$ of positions of adjacent nonzero digits
3. Construct the linear form $\Lambda = -n\log a + (m-m(i))\log b + \log q$
4. Apply Baker's theorem to obtain a lower bound on $|\Lambda|$
5. Through a chain of inequalities, derive $m(i+1)/m(i) < C\log n$
6. Sum over all ratios to obtain the final result

Experimental Setup

Note: This is a pure theoretical mathematics paper with no experimental verification. This section describes numerical illustrations and theoretical validations presented in the paper.

Numerical Illustrations

The paper clarifies concepts through concrete examples:

1. Sequence $2^n$ (OEIS A000079):
   • First 11 terms: 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 1024, ...
2. Digit Sum Sequence (OEIS A001370):
   • Corresponding digit sums: 1, 2, 4, 8, 7, 5, 10, 11, 13, 8, 7, ...
   • Demonstrates non-monotonicity
3. Illustrative Diagram (Figure 1):
   • $2^{103} = 10141204801825835211973625643008$
   • Partition digits into blocks: 10141204801825835 | 2119736256 | 43008
   • Each block contains at least one nonzero digit

Theoretical Validation Methods

1. Mathematical Induction: Proofs of Theorems 1-3 employ mathematical induction
2. Constructive Proof: Existence is established through explicit construction of the sequence $e(k)$
3. Asymptotic Analysis: Growth rates are analyzed using big-O and Theta notation

Exercises

The paper provides two exercises for readers to verify understanding:

Exercise 1: Prove that every power of 3 has a multiple $m$ (not divisible by 10) such that $c_{10}(m) = 2$.

Exercise 2: Prove that the count of nonzero decimal digits of the $n$-th Fibonacci number tends to infinity.

Experimental Results

Summary of Main Theoretical Results

Comparison with Existing Results

1. Senge-Straus (1973):
   • Result: $c_b(a^n) \to \infty \Leftrightarrow \log(a)/\log(b)$ irrational
   • Improvement in this paper: Provides explicit logarithmic lower bound
2. Stewart (1980):
   • Result: $c_b(a^n) > \frac{\log n}{\log\log n + C} - 1$ (general conditions)
   • Relation to this paper: Theorem 8 reproduces this result; Theorem 4 provides stronger bounds under more restrictive conditions
3. Sanna (2015):
   • Result: $s_b(n!) > C\log n\log\log\log n$
   • Relation to this paper: Theorem 5 provides a weaker but more elementary bound $c_b(n!) > C\log n$

Methodological Advantages

Key Findings

1. Power of the $\xi$ Function: The correction function $\xi$ elegantly handles the trailing zeros problem and is key to method generalization.
2. Essence of Irrationality Condition:
   • $\log(d)/\log(b)$ irrational is equivalent to $f_i/e_i$ not all equal
   • This ensures $\xi(a) > 0$, thus $\xi(a^n)$ grows linearly
3. Sequence-Specific Behavior:
   • Factorials: $\xi(n!) = \Theta(n)$ → $c_b(n!) = \Theta(\log n)$
   • LCM: $\xi(\Lambda_n) = \Theta(\log n)$ → $c_b(\Lambda_n) = \Theta(\log\log n)$
   • Reflects intrinsic structural differences among sequences
4. Necessity: If $\log(a)/\log(b) = r/s \in \mathbb{Q}$, then $a^{ns} = b^{nr}$ has only one nonzero digit, showing the irrationality condition is necessary.

Related Work

Historical Development

1. Sierpiński (1970):
   • Posed the problem of proving that the decimal digit sum of $2^n$ tends to infinity
   • Initiated the classical problem of digit sum research
2. Senge & Straus (1973):
   • First provided the necessary and sufficient condition: $c_b(a^n) \to \infty \Leftrightarrow \log(a)/\log(b)$ irrational
   • Employed PV-number (Pisot-Vijayaraghavan number) theory
   • Did not provide quantitative bounds on growth rate
3. Baker (1975):
   • Developed transcendental number theory of linear forms in logarithms
   • Provided effective lower bounds, becoming an important tool for subsequent research
4. Stewart (1980):
   • First provided quantitative bounds: $c_b(a^n) > \frac{\log n}{\log\log n + C} - 1$
   • Used Baker's theorem in the proof
   • The method is quite technical and difficult to understand
5. Sanna (2015):
   • Extended research to factorials and LCM
   • Proved $s_b(n!) > C\log n\log\log\log n$
   • Employed prime number theorem and refined number-theoretic estimates

Related Research Directions

1. Normality of Digit Sums:
   • Studies distribution of digits in various bases
   • Conjecture: Decimal digit sum of $2^n$ is approximately $4.5n\log_{10}2$ (unproven)
2. Digit Sums of Other Sequences:
   • Fibonacci numbers (addressed in Exercise 2)
   • Prime powers
   • Polynomial values
3. Higher-Dimensional Generalizations:
   • Powers of multiple variables
   • Multi-base representations
4. Computational Complexity:
   • Algorithm efficiency for computing digit sums
   • Connections to automata theory

Positioning of This Work

The unique contributions of this paper are:

1. Methodological Innovation: Systematically develops elementary methods based on divisibility, bridging the gap between elementary methods and advanced tools.
2. Unified Framework: Through the $\xi$ function, establishes a unified treatment framework applicable to multiple sequence types.
3. Pedagogical Value: Provides a clear path from concrete problems to general theory, suitable for teaching and learning.
4. Result Improvement: Under specific conditions, achieves stronger bounds than Stewart ($\log n$ vs. $\frac{\log n}{\log\log n}$).

Conclusions and Discussion

Main Conclusions

1. Core Theorem: Under the condition that $\log(d)/\log(b)$ is irrational, the count of nonzero digits of $a^n$ in base $b$ grows at least at the rate $C\log n$.
2. Broad Applicability: The method applies not only to power sequences but also to factorials ($\log n$ growth) and LCM ($\log\log n$ growth).
3. Elementary Nature: All results in Sections 1-4 use only elementary number theory, requiring no transcendental number-theoretic tools.
4. Completeness: Section 5 employs Baker's theorem to provide a complete proof for the most general case, reproducing the best-known results.

Limitations

1. Condition Restrictions:
   • Theorem 4 requires $\log(d)/\log(b)$ irrational, which is stronger than Theorem 8's condition ($\log(a)/\log(b)$ irrational)
   • Example: For $a=6, b=10$, we have $d=2$, and $\log(2)/\log(10)$ is irrational, so Theorem 4 applies
   • But for $a=15, b=10$, we have $d=3$, and $\log(3)/\log(10)$ is irrational, though this may not be the optimal condition
2. Bound Strength:
   • For factorials, the bound $c_b(n!) > C\log n$ is weaker than Sanna's $s_b(n!) > C\log n\log\log\log n$
   • The cost of elementary methods is obtaining weaker bounds
3. Non-Explicit Constants:
   • While existence of constant $C > 0$ is proved, no explicit expression for $C$ is provided
   • May be inconvenient for practical applications
4. Missing Upper Bounds:
   • Paper focuses on lower bounds without discussing upper bounds
   • For example, does $c_b(a^n) = O(n)$ hold?
5. Digit Sum vs. Nonzero Digit Count:
   • Main results concern $c_b(n)$ (count of nonzero digits)
   • While asymptotically equivalent to $s_b(n)$ (digit sum), constant factors may be important

Future Directions

1. Improved Bounds:
   • Can elementary methods achieve $c_b(a^n) = \Omega(\log n \log\log n)$?
   • Can the gap with Sanna's results be narrowed?
2. Explicit Constants:
   • Compute explicit expressions for constant $C$
   • Provide precise estimates for small values of $a, b$
3. Generalization to Other Sequences:
   • Fibonacci numbers (suggested by Exercise 2)
   • Catalan numbers
   • Prime sequences
4. Digit Distribution:
   • Prove or disprove the uniform digit distribution conjecture
   • Study asymptotic formulas for digit sums
5. Computational Applications:
   • Develop efficient algorithms for computing digit sums
   • Applications to cryptography and coding theory
6. Multidimensional Generalizations:
   • Study digit sums of numbers of the form $a^m b^n$
   • Mixed-radix representations

In-Depth Evaluation

Strengths

1. Methodological Innovation

• Combination of Elementarity and Depth: Successfully solves problems seemingly requiring advanced tools using purely elementary methods, demonstrating the power of elementary number theory.
• Unified Framework: The introduction of the $\xi$ function is a clever innovation, elegantly handling the trailing zeros problem and enabling broad applicability.
• Constructive Nature: Proofs are completely constructive, allowing in principle explicit bounds for any $n$.

2. Theoretical Contributions

• Quantitative Improvement: Under specific conditions, improves from $\frac{\log n}{\log\log n}$ to $\log n$; while conditions are stronger, the bounds are better.
• Generalizability: First unified elementary treatment of three sequence classes: powers, factorials, and LCM.
• Completeness: Provides both elementary proofs and applications of Baker's theorem, offering a complete theoretical picture.

3. Writing Quality

• Clear Structure: Progresses from special to general, concrete to abstract, with logical clarity.
• Intuitive Guidance: Uses concrete examples like Figure 1 to aid understanding.
• Teaching-Oriented: Includes exercises, suitable for pedagogical use.
• Historical Context: Thoroughly introduces problem history and related work.

4. Technical Details

• Rigor: All theorems have complete proofs without gaps.
• Boundary Handling: Carefully addresses various boundary cases ($k=1$, trailing zeros, etc.).
• Consistent Notation: Introduced symbols ($c_b, s_b, \nu_p, \xi$) are clear and consistent.

Weaknesses

1. Theoretical Limitations

• Condition Strength: Theorem 4's conditions are stronger than Theorem 8's, limiting applicability.
  • Example: For $a=15, b=10$, need to verify $\log(3)/\log(10)$ is irrational.
• Suboptimal Bounds: Bounds for factorials are weaker than best-known results.
  • This paper: $c_b(n!) > C\log n$
  • Sanna: $s_b(n!) > C\log n\log\log\log n$
• Missing Upper Bounds: No discussion of upper bounds for $c_b(a^n)$, leaving the theoretical picture incomplete.

2. Technical Details

• Hidden Constants: Constant $C$ depends on $a, b$ but no explicit expression is provided, inconvenient for applications.
• Asymptotic Notation: Frequent use of $\Theta, O, o$ notation, while concise, sometimes obscures precise relationships.
• $\xi$ Function Selection: Definition of $\xi$ depends on choice of primes $p, q$; different choices may yield different bounds, insufficiently discussed.

3. Proof Techniques

• Non-Constructive Induction: While proofs are constructive, the inductive process makes actual computation of $C$ difficult.
• Baker's Theorem as Black Box: Section 5 uses Baker's theorem as a "black box," contrasting with earlier elementarity, though the author clearly acknowledges this.

4. Practical Utility

• Computational Efficiency: No discussion of algorithm efficiency for computing $c_b(a^n)$.
• Numerical Verification: Lacks concrete numerical examples verifying tightness of theoretical bounds.
• Application Scenarios: Does not discuss practical applications (e.g., cryptography, coding theory).

Impact Assessment

Contribution to the Field

1. Methodological Contribution: Provides new elementary tools for digit sum problems, potentially inspiring research on related problems.
2. Teaching Resource: Serves as excellent teaching material, demonstrating how to develop profound theory from simple problems.
3. Bridge Role: Connects elementary methods and advanced tools (Baker's theorem), providing entry points for researchers with different backgrounds.

Practical Value

1. Theory-Focused: Primarily pure mathematics theory with limited direct practical utility.
2. Potential Applications:
   • Analysis of pseudorandom number generators
   • Study of digit properties in cryptography
   • Computational complexity theory

Reproducibility

1. Fully Reproducible: All proofs are complete, allowing readers to verify step-by-step.
2. Easy Implementation: Methods based on divisibility are straightforward to program.
3. Learning Support: Provided exercises help readers consolidate understanding.

Applicable Scenarios

1. Academic Research

• Number Theorists: Provides new technical tools applicable to related problems.
• Combinatorics: Digit sum problems have deep connections to combinatorial structures.
• Computational Number Theory: Provides theoretical foundations for algorithm design.

2. Teaching Contexts

• Upper-Level Undergraduate/Graduate Courses: Excellent case study in number theory pedagogy.
• Mathematical Competitions: Sierpiński's problem is suitable as competition material.
• Science Communication: Exemplifies progression from simple problems to profound theory.

3. Subsequent Research

• Generalization Directions: Provides template for studying digit sums of other sequences.
• Improvement Directions: Provides foundation for seeking stronger bounds.
• Interdisciplinary Connections: May connect with dynamical systems and ergodic theory.

Overall Assessment

This is an excellent pure mathematics paper with the following outstanding characteristics:

1. Theoretical Depth: Despite using elementary methods, achieves meaningful new results.
2. Methodological Innovation: Introduction of the $\xi$ function and unified framework represent genuine innovation.
3. Writing Quality: Clear, rigorous, and pedagogically rich—exemplary mathematical writing.
4. Completeness: Offers both elementary proofs and advanced tool applications, providing complete theoretical landscape.

Primary Value:

• For number theorists: Provides new tools
• For educators: Provides excellent teaching material
• For students: Provides learning pathway

Main Shortcomings:

• Bounds are not optimal in some cases
• Lacks explicit constants and numerical verification
• Limited practical applicability

Recommendation Rating: ⭐⭐⭐⭐☆ (4.5/5)

• Strongly recommended for number theorists and students
• Limited value for applied researchers

References

Key references cited in the paper:

1. Andrica et al. (2020): Exponential properties in group theory, providing theoretical foundation for LCM.
2. Baker (1975): Transcendental Number Theory, classic textbook in transcendental number theory, source of Baker's theorem.
3. Dickson (1919): History of the Theory of Numbers, classic in number theory history, contains Legendre's formula.
4. Sanna (2015): "On the sum of digits of the factorial," best-known result for factorial digit sums.
5. Senge & Straus (1973): "PV-numbers and sets of multiplicity," first to provide necessary and sufficient conditions.
6. Sierpiński (1970): 250 Problems in Elementary Number Theory, original source of the problem.
7. Stewart (1980): "On the representation of an integer in two different bases," first to provide quantitative bounds.

Summary: Through clever elementary methods, this paper achieves meaningful progress on the classical digit sum problem, combining theoretical depth with pedagogical value—an excellent contribution to number theory.