Baker's Theorem

In the vast landscape of number theory, certain deep results act as master keys, unlocking problems that have remained sealed for centuries. Baker's theorem is one such monumental achievement, offering a profound insight into the very fabric of numbers. For much of the 20th century, a significant gap existed in our ability to solve many ancient mathematical puzzles known as Diophantine equations; while theorems by Thue, Siegel, and Roth proved that solutions were finite, they provided no way to actually find them. This article navigates the revolutionary impact of Alan Baker's work, which bridged this gap by introducing the concept of "effectivity." In what follows, we will first explore the principles and mechanisms behind the theorem, demystifying the concept of linear forms in logarithms and the quantitative bound that lies at the theorem's heart. We will then journey through its far-reaching applications, from taming infinite solution sets of famous equations to building surprising bridges with algebraic geometry. Our exploration begins by examining the core question that Baker's theorem so powerfully answers: if a special combination of logarithms is not zero, exactly how close to zero can it be?

Imagine you are standing on a number line, a simple, straight road stretching to infinity in both directions. Now, imagine a special point on this line: zero. For centuries, mathematicians have been fascinated by questions about zero. A simple question like "when is $ax - b = 0$?" has a simple answer: when $x = b/a$. But what if the stage for our questions is not so simple? What if our numbers are not just integers or rationals, but more exotic creatures, and our equations are not simple lines but intricate tapestries woven from logarithms? This is the world of Baker's theorem, a world where the seemingly simple question, "Can this number be zero?", and its more subtle cousin, "If it's not zero, how close to zero can it be?", unlock profound secrets about the very nature of numbers.

The central character in our story is a quantity called a ​​linear form in logarithms​​. It looks innocent enough: $\Lambda = b_1 \log \alpha_1 + b_2 \log \alpha_2 + \cdots + b_n \log \alpha_n$ Here, the $b_i$ are simple integers, and the $\alpha_i$ are ​​algebraic numbers​​—numbers that are roots of polynomial equations with rational coefficients, like $\sqrt{2}$ or the golden ratio $\phi$. The trouble, and the beauty, lies in that little word "log". This is not the familiar logarithm from high school. This is the ​​complex logarithm​​.

You might think you know what a logarithm is, but wait until you see it in the complex plane. For a positive real number, say $x$, the logarithm $\ln(x)$ is a unique real number. But for a complex number, things get wonderfully strange. A complex number like $z$ can be described by its distance from the origin, $|z|$, and its angle, $\theta$. Its logarithm turns out to be $\ln|z| + i\theta$. But which angle? The angle $\theta$ is the same as $\theta + 2\pi$, or $\theta + 4\pi$, and so on. This means every complex number (except zero) has not one logarithm, but an infinite, evenly spaced ladder of them, each value differing from the next by a multiple of $2\pi i$. For example, the number $-1$ has logarithms $i\pi$, $3i\pi$, $-i\pi$, and so on, forever.

This presents a problem. If each $\log \alpha_i$ in our form $\Lambda$ can be any of an infinite set of values, then $\Lambda$ itself is not a single number but a whole constellation of possibilities. To do any meaningful mathematics, we must first agree to tame this ambiguity. The standard procedure is to make a specific choice for each logarithm. Most commonly, we choose the ​​principal branch​​, where the angle is restricted to the interval $(-\pi, \pi]$. By fixing a branch for each $\log \alpha_i$, we ensure that our linear form $\Lambda$ becomes a single, well-defined complex number, a specific point in the complex plane. This choice is not a mere technicality; it is the essential first step to asking any sensible question about the value of $\Lambda$.

Now that we have our well-defined number $\Lambda$, we can ask the first great question: Can it be zero?

The answer is, sometimes. Let's see how. Remember the wonderful property of the exponential function, $\exp(x+y) = \exp(x)\exp(y)$. Applying this to our form $\Lambda$ gives: $\exp(\Lambda) = \exp(b_1 \log \alpha_1 + \cdots + b_n \log \alpha_n) = \alpha_1^{b_1} \alpha_2^{b_2} \cdots \alpha_n^{b_n}$ Now, the other key property of the exponential function is that $\exp(z) = 1$ if and only if $z$ is an integer multiple of $2\pi i$. So, if we ever find that the product $\alpha_1^{b_1} \cdots \alpha_n^{b_n} = 1$, it means that our linear form $\Lambda$ must be an integer multiple of $2\pi i$. This is a beautiful bridge between the multiplicative structure of the numbers $\alpha_i$ and the additive structure of their logarithms.

When there exists a non-trivial set of integers $b_i$ such that $\alpha_1^{b_1} \cdots \alpha_n^{b_n} = 1$, we say the numbers $\alpha_1, \dots, \alpha_n$ are ​​multiplicatively dependent​​. In this case, it is possible for a corresponding linear form in their logarithms to vanish (modulo $2\pi i$).

But what if they are ​​multiplicatively independent​​? What if no such relation exists? In that case, the famous Gelfond-Schneider theorem gave a preliminary, qualitative answer for a simple case: a form like $\beta_1 \log \alpha_1 + \beta_2 \log \alpha_2$ (with algebraic coefficients) cannot be zero unless there's a good reason (like a rational relationship between the coefficients). This result implied, for example, that $2^{\sqrt{2}}$ must be transcendental. It was a wonderful result, but it was qualitative. It said $\Lambda$ is not zero, but it didn't say anything more.

This leads us to the deeper, more subtle question. If $\Lambda$ cannot be zero, can it get arbitrarily close to zero? The answer is yes! Just as we can approximate an irrational number like $\pi$ with fractions ($22/7$, $355/113$, etc.) to astonishing accuracy, we can always find clever choices of large integers $b_i$ to make the value of $\Lambda$ tantalizingly close to zero. The real question, the one that lies at the heart of modern number theory, is not if $\Lambda$ can be small, but how small, as a function of the size of the integers $b_i$ we use to construct it.

This is where Alan Baker entered the scene and changed the landscape forever. Baker's theorem provides a powerful, explicit answer to the question "how small?". It provides a ​​lower bound​​ for $|\Lambda|$. It builds a fence around zero and says that no non-zero value of $\Lambda$ can ever enter this forbidden region.

Qualitatively, the theorem states that if $\Lambda \neq 0$, then there is an effectively computable constant $C > 0$ such that: $|\Lambda| > B^{-C}$ where $B$ is a measure of the size of the integer coefficients, for instance $B = \max |b_i|$. The constant $C$ depends on the number of terms, $n$, and the complexity (degree and height) of the algebraic numbers $\alpha_i$.

Why is this so revolutionary? Let's contrast it with the kind of bounds that came before, known as Liouville-type bounds. A classical approach would be to look at the number $\beta = \exp(\Lambda) = \prod \alpha_i^{b_i}$. If $\Lambda$ is close to zero, $\beta$ must be close to one. Liouville's methods could provide a lower bound for $|\beta - 1|$, which translates into a lower bound for $|\Lambda|$. However, this bound would be incredibly weak, something on the order of $\exp(-C'B)$. This decays exponentially in $B$. The difference between a bound of $B^{-C}$ (polynomial decay) and $\exp(-C'B)$ (exponential decay) is colossal. It's the difference between a leaky faucet and a waterfall. The polynomial-decay bound of Baker is exponentially stronger. It says that while $\Lambda$ can get small as the coefficients $B$ grow, it cannot do so "too quickly". This quantitative precision, this control over the rate of approach to zero, is what makes the theorem a true "measure of linear independence" over the rationals and, as it turns out, over all algebraic numbers.

The most magical word in the description of Baker's theorem is ​​effective​​. This means that the constant $C$ in the inequality is not just some abstract entity that we know exists; it is computable. Given the $\alpha_i$, we can, in principle, sit down and calculate a specific number for the bound.

This is a profound distinction from many other powerful results in number theory, such as Roth's theorem. Roth's theorem gives the best possible qualitative statement about how well algebraic numbers can be approximated by rationals, but it is ​​ineffective​​. It's like an oracle that tells you there are only a finite number of needles in a haystack but gives you no clue how big the haystack is. You can't use it to find the needles. Baker's theorem, by being effective, tells you the size of the haystack. It gives you a concrete, finite region to search for solutions.

How does this work in practice? Consider a famous type of Diophantine equation, the ​​S-unit equation​​, like $x+y=1$, where we are looking for solutions $x$ and $y$ that are built from a fixed, finite set of prime numbers. If a solution $(x,y)$ were to exist with enormous integer components, then one of the terms, say $y$, must be very small. This forces $x$ to be very close to $1$. But $x$ being an S-unit means $\log x$ is a linear form in the logarithms of our fixed primes with integer coefficients. If $x$ is close to $1$, then $\log x$ must be close to $0$.

Here's the master stroke. The Diophantine equation gives us an upper bound on our linear form $|\Lambda| = |\log x|$, which gets smaller and smaller as the hypothetical solution gets larger. Baker's theorem, on the other hand, gives us a concrete lower bound for $|\Lambda|$. For a large enough hypothetical solution, the upper bound from the equation will crash through the lower bound set by Baker's theorem, creating a contradiction. This proves that no solution larger than a certain explicitly computable size can exist. The seemingly unsolvable problem of an infinite search is reduced to a finite, manageable one. This crucial link often involves a simple but vital lemma relating $|\Lambda|$ to $|\exp(\Lambda)-1|$: for small $\Lambda$, the two are roughly proportional, with $|e^{\Lambda}-1| \ge |\Lambda|/2$ providing a rigorous bridge.

One final piece of the puzzle demonstrates the beautiful unity of algebra that underlies this theory. The core theorem gives bounds for linear forms with integer coefficients $b_i$. But what if we want to study a form where the coefficients $\beta_i$ are themselves algebraic numbers, like $L = \sqrt{2} \log 3 - \sqrt{5} \log 7$?

The genius of the method is to use the structure of algebraic number fields. Any field of algebraic numbers, say $K = \mathbb{Q}(\beta_1, \dots, \beta_n)$, can be viewed as a finite-dimensional vector space over the rational numbers $\mathbb{Q}$. This means we can pick a basis, say $\{w_1, \dots, w_m\}$, and write every coefficient $\beta_i$ as a unique combination of these basis elements with rational coefficients.

By substituting these expressions back into our linear form $L$ and rearranging the sums, we can transform our single linear form with algebraic coefficients into a set of $m$ simultaneous linear forms, each with integer coefficients. A fundamental theorem of algebra guarantees that these new forms are related to our original form $L$ via an invertible matrix built from the embeddings of the field $K$ into the complex numbers.

This incredible maneuver means that any question about the original form $L$—whether it is zero, or how small it can be—is perfectly translated into a question about the collection of simpler, integer-coefficient forms. It's like using an algebraic lever to break down one complex problem into several simpler ones that we already know how to handle. It is a stunning display of how the abstract structures of modern algebra provide concrete tools for solving ancient problems about numbers, revealing the deep and elegant interconnectedness of mathematics.

Having peered into the intricate machinery of Baker's theorem in the previous chapter, you might be left with a sense of wonder, but also a practical question: What is it good for? A beautiful theorem is one thing, but a useful one is another. It's like being shown a marvelously crafted microscope; the real joy comes when you use it to see the world anew. Baker's theorem is just such a device. It is not merely a statement of fact, but a powerful lens that reveals a hidden, rigid structure in the seemingly chaotic world of numbers.

In this chapter, we take this new lens out into the field. We'll see how it acts as a master key, unlocking Diophantine puzzles that had stumped mathematicians for centuries. We'll watch it build bridges between the discrete world of number theory and the continuous landscapes of algebraic geometry. And we'll use it as a ruler to take the measure of numbers themselves, quantifying their deepest properties. This is the story of how a profound insight into the nature of logarithms radiates outward to illuminate vast tracts of the mathematical universe.

One of the great quests in number theory is to classify numbers: are they algebraic, like $\sqrt{2}$, or are they transcendental, like $\pi$? For a long time, proving a number was transcendental was an act of profound difficulty. A major victory came when the Gelfond–Schneider theorem solved Hilbert's seventh problem, establishing that for any algebraic number $\alpha \neq 0, 1$ and any irrational algebraic number $\beta$, the value $\alpha^{\beta}$ is transcendental.

Baker's method provides a stunningly different and more powerful proof of this fact. The original proofs were clever, but Baker's approach says something deeper. It doesn't just show that $\alpha^\beta$ is transcendental; it shows that it cannot be too well approximated by other algebraic numbers. The argument is a beautiful example of a proof by contradiction. You start by assuming the opposite: suppose $\gamma = \alpha^\beta$ is algebraic. From this assumption, you can construct a very special number, a linear form in logarithms such as $\Lambda = \beta \log \alpha - \log \gamma$. Because of the way logarithms work, this number would have to be incredibly, ridiculously close to zero. Yet, it can't be exactly zero, because that would imply $\beta$ is rational, which we assumed it isn't. Here is the trap: Baker's theorem provides a rigorous, non-negotiable floor for how small such a nonzero linear form can be. The upper bound you derive from your assumption turns out to be smaller than the minimum possible size dictated by Baker's theorem. It’s like proving a creature can't exist because it would have to be smaller than its own atoms. The contradiction is inescapable, and the only way out is to discard the initial assumption. Thus, $\alpha^{\beta}$ must be transcendental.

This is powerful, but where does it lead? The Gelfond–Schneider theorem is about a single number. What if we have a collection of such numbers, like $2^{\sqrt{2}}, 3^{\sqrt{3}}, 5^{\sqrt{5}}, \dots$? Are they related to each other in some hidden polynomial way? In other words, are they algebraically independent? The Gelfond–Schneider theorem is silent on this point. It's easy to construct examples where numbers of this form are individually transcendental but algebraically dependent; for instance, $2^{\sqrt{2}}$ and $2^{2\sqrt{2}} = (2^{\sqrt{2}})^2$ are clearly dependent. To make any headway on the general question of algebraic independence, one needs the full force of Baker's theory, which deals with linear forms in many logarithms and provides the first and most crucial tools for exploring these deeper, collective relationships among transcendental numbers.

Perhaps the most celebrated application of Baker's theorem lies in the solution of Diophantine equations—polynomial equations for which we seek integer solutions. For millennia, these stood as individual puzzles, each requiring its own unique flash of insight. The work of Thue, Siegel, and Roth in the 20th century showed that many important classes of these equations have only a finite number of solutions. This was a revolutionary discovery, but it came with a frustrating catch: the proofs were "ineffective." They proved finiteness by contradiction but gave no method, not even in principle, to find the solutions or even to put an upper bound on their size. The solutions were finite, but they were lost in an infinite sea.

Baker's method changed everything. It provided the first effective bounds. By converting these equations into problems about linear forms in logarithms, it turned an infinite search into a finite, and in principle completable, one.

A classic first stop is the deceptively simple ​​$S$-unit equation​​, $x+y=1$. If we restrict the solutions $x$ and $y$ to be a special type of number called $S$-units—numbers built from a finite list of prime factors—their structure is rigidly controlled. The equation forces a delicate cancellation between $x$ and $y$ that, through the magic of logarithms, translates into a linear form being very close to zero. Baker’s theorem puts a stop to this, bounding the size of any possible solutions and making them countable. This seemingly niche equation is a gateway, as we will see, to solving problems on much more complex geometric objects.

A more famous puzzle is ​​Catalan's conjecture​​, which asks for all integer solutions to $x^m - y^n = 1$. For over a century, the only known solution with all variables greater than one was $3^2 - 2^3 = 1$. Were there others? The problem seemed infinite. By rewriting the equation and taking logarithms, one can create the linear form $\Lambda = m \ln x - n \ln y = \ln(1+y^{-n})$. For large solutions, $\Lambda$ becomes vanishingly small. Baker's method, by providing a lower bound on $\Lambda$ in terms of the exponents $m$ and $n$, allowed Robert Tijdeman in 1976 to prove that there is an absolute, computable upper bound on the size of any possible solution. The bound was astronomically large, far too big for a computer search, but it was a staggering achievement: an infinite problem had been reduced to a finite one. (The puzzle was fully solved in 2002 by Preda Mihăilescu using entirely different, algebraic methods—a beautiful illustration of how different streams of thought can converge on the same truth.)

This same principle applies to a vast family of equations, most notably ​​Thue equations​​ like $F(x,y) = m$, where $F(x,y)$ is an irreducible polynomial of degree at least 3. Through the machinery of algebraic number theory, solving such an equation can be transformed into a problem about units in a number field. These units, much like the $S$-units above, have a structure that can be described by a finite set of generators. The Thue equation forces a relationship between these units that once again manifests as a linear form in logarithms being extremely close to zero. Baker's theorem provides the effective bound on the exponents of the fundamental units, and from there, an effective bound on the size of the integer solutions $|x|$ and $|y|$. This can be extended even further to ​​Thue-Mahler equations​​, where the right-hand side can also involve prime factors from a given set. For the first time, we had a systematic algorithm to find all integer solutions to a wide class of ancient problems.

The power of these methods extends beyond discrete equations into the world of geometry. Finding integer solutions to an equation like $y^2 = x^3 + 17$ is equivalent to finding points on the corresponding curve that have integer coordinates. These are called "integral points."

For certain classes of curves, the problem of finding all their integral points can be brilliantly reduced to solving an $S$-unit equation. This is the case for curves of genus 0 with at least three points "at infinity" (think of the line with three points removed) and for curves of genus 1, known as elliptic curves. In these situations, one can construct special functions on the curve that map any integral point to a solution of the $S$-unit equation $u+v=1$. Since Baker's theory gives us an effective method to find all solutions to the $S$-unit equation, we can work backward to find all the integral points on our original curve.

This forms a beautiful intellectual bridge: a geometric problem of points on a curve is translated into an algebraic problem about $S$-units, which is then solved by an analytic tool from transcendental number theory—Baker's theorem. Here, we see the profound unity of mathematics in action. This effectiveness stands in sharp contrast to the general case for curves of genus 2 or higher. While Siegel's theorem guarantees that these curves also have only a finite number of integral points, the proof is ineffective, leaving us with no algorithm to find them. Baker's method thus illuminates exactly where our ability to compute effectively currently begins and ends, drawing a sharp line between the tractable and the mysterious.

Finally, let's return to the numbers themselves. Baker's theorem is fundamentally quantitative. It doesn't just say a form is non-zero; it says how non-zero it must be. This allows us to "measure" certain properties of numbers.

One such property is the ​​irrationality measure​​, which quantifies how well a number can be approximated by fractions. To find a bound on the irrationality measure of $\ln 2$, we would study how close $|\ln 2 - p/q|$ can get to zero. This is equivalent to studying the smallness of the linear form $|q \ln 2 - p|$. Baker's theorem gives us an explicit lower bound on this quantity, which translates directly into an explicit upper bound for the irrationality measure of $\ln 2$. We get a concrete, computable handle on the transcendental nature of this number.

In the same spirit, Baker's theorem quantifies the notion of ​​multiplicative independence​​. For a set of integers like $\{6, 10, 15\}$, prime factorization tells us that the only way $6^x 10^y 15^z = 1$ is if $x,y,z$ are all zero. This is equivalent to saying the linear form $x \ln 6 + y \ln 10 + z \ln 15$ is zero only for the trivial solution. But Baker's theorem goes further: it gives a lower bound on how far from 1 any other product $6^x 10^y 15^z$ must be. It provides a measure of repulsion from the point 1.

From proving the impossibility of certain numbers, to taming the infinite solutions of ancient equations, to charting the geometric landscape of curves, and finally to fashioning a ruler for the number line itself, the applications of Baker's theorem are as profound as they are diverse. It stands as a monumental testament to how a single, deep insight into the structure of numbers can resonate across the entire landscape of mathematics, revealing hidden connections and a beautiful, underlying order.