Power Sum Polynomials

In mathematics, a collection of numbers can be described in fundamentally different ways. We can focus on the properties of the individuals, such as the sum of their squares or cubes, or we can examine their collective interactions, like the sum of all pairwise products. These two perspectives give rise to two crucial families of functions: the power sum symmetric polynomials ($p_k$) and the elementary symmetric polynomials ($e_k$). While they appear to capture distinct information, they are in fact intimately related, and understanding this connection unlocks a powerful and unifying mathematical tool. This article addresses the apparent gap between these two descriptions, revealing the elegant machinery that translates between them.

This article will guide you through this profound connection. In the first section, "Principles and Mechanisms," we will explore the definitions of both polynomial families and derive the foundational equations, known as Newton's Identities, that link them. In the second section, "Applications and Interdisciplinary Connections," we will witness the remarkable utility of this relationship, seeing how it provides computational shortcuts in algebra and linear algebra and how the same structural pattern echoes through advanced fields like geometry, topology, and even theoretical physics.

Imagine you are looking at a crowd of people. You could describe this crowd in two fundamentally different ways. First, you could focus on the individuals. You might note the height of each person, their age, or some other personal attribute. If you wanted to be mathematical, you could take some measurement for each person—say, their height $x_i$—and then compute the sum of their heights, $x_1 + x_2 + \dots + x_n$, or the sum of the squares of their heights, $x_1^2 + x_2^2 + \dots + x_n^2$. This is one way of capturing information about the group.

Alternatively, you could describe the crowd by its collective interactions. You could look at pairs of people and the relationships between them, or groups of three, and so on. This approach isn't about individual attributes, but about the structure of the group as a whole.

In the world of mathematics, these two perspectives correspond to two families of remarkable objects: the ​​power sum symmetric polynomials​​ ($p_k$) and the ​​elementary symmetric polynomials​​ ($e_k$). They are the bedrock of the theory of symmetric functions, and understanding their relationship is like discovering a secret passage that connects seemingly distant mathematical lands.

Let's take a small set of variables, say $x_1, x_2, x_3$.

The ​​power sum polynomials​​, denoted $p_k$, are the embodiment of the first perspective: focusing on individual attributes and summing them up. We simply take the $k$-th power of each variable and add them together. $p_1 = x_1^1 + x_2^1 + x_3^1$ $p_2 = x_1^2 + x_2^2 + x_3^2$ $p_3 = x_1^3 + x_2^3 + x_3^3$ And so on. There's a beautiful simplicity to them. $p_k$ is the sum of the $k$-th powers.

The ​​elementary symmetric polynomials​​, denoted $e_k$, represent the second perspective: the structure of collective interactions. We build them by taking all possible products of the variables, grouped by size.

• $e_1$ is the sum of all variables taken one at a time: $e_1 = x_1 + x_2 + x_3$
• $e_2$ is the sum of all products of variables taken two at a time: $e_2 = x_1x_2 + x_1x_3 + x_2x_3$
• $e_3$ is the sum of all products of variables taken three at a time: $e_3 = x_1x_2x_3$

Notice that for three variables, you can't choose four distinct variables, so $e_k$ for $k > 3$ is zero. You might also notice that $p_1$ and $e_1$ are identical. This is the first hint of a connection. But are they otherwise related? Can one family be described in terms of the other?

Let's try a little experiment. What happens if we square $p_1$ (which is the same as $e_1$)? For any number of variables $x_1, \dots, x_n$: $p_1^2 = (x_1 + x_2 + \dots + x_n)^2 = (x_1 + x_2 + \dots + x_n)(x_1 + x_2 + \dots + x_n)$

When we expand this, we get two kinds of terms. We get terms where a variable is multiplied by itself, like $x_1^2, x_2^2, \dots$. And we get cross-terms, where one variable is multiplied by a different one, like $x_1x_2, x_2x_1$, and so on.

Let's gather them up. The sum of the squared terms is simply: $x_1^2 + x_2^2 + \dots + x_n^2$ But this is just our friend, the second power sum, $p_2$!

What about the cross-terms? We get every product $x_ix_j$ where $i \neq j$. Furthermore, for each pair, say $x_1$ and $x_2$, we get both $x_1x_2$ from the expansion and $x_2x_1$. Since multiplication doesn't care about order, this is just $2x_1x_2$. So the sum of all the cross-terms is exactly twice the sum of all products of distinct pairs of variables. $2 \sum_{1 \le i < j \le n} x_i x_j$ And the sum inside this expression is precisely the definition of the second elementary symmetric polynomial, $e_2$.

Putting it all together, we've stumbled upon a remarkable identity: $p_1^2 = p_2 + 2e_2$

Just by squaring the simplest sum, we have uncovered a rigid, fundamental relationship between the first two power sums and the second elementary symmetric polynomial. We can rearrange this to express $e_2$ using only power sums: $e_2 = \frac{1}{2}(p_1^2 - p_2)$

This isn't just a neat trick; it's a crack of light that reveals a deep and intricate structure. It tells us that these two different ways of looking at a collection of variables are not independent at all. They are intrinsically linked.

This simple bridge is just the beginning. The full relationship between the two families of symmetric polynomials is captured by a magnificent set of equations known as ​​Newton's Identities​​ (or Newton's Sums). They act as a "Rosetta Stone," allowing us to translate perfectly between the language of $p_k$ and the language of $e_k$.

These identities form a recursive ladder. If you know all the elementary polynomials $e_1, \dots, e_k$, you can climb the ladder to find any power sum $p_k$. Conversely, if you know the power sums $p_1, \dots, p_k$, you can find any $e_k$.

The identities are as follows (with $e_0 = 1$ by convention): $p_1 - e_1 = 0$ $p_2 - e_1 p_1 + 2e_2 = 0$ $p_3 - e_1 p_2 + e_2 p_1 - 3e_3 = 0$ $\vdots$ In general, for $k \le n$ (the number of variables): $p_k - e_1 p_{k-1} + e_2 p_{k-2} - \dots + (-1)^{k-1} e_{k-1} p_1 + (-1)^k k e_k = 0$

You can test these yourself. For instance, if you take $x_1=1, x_2=2, x_3=0$, you can calculate $p_3 = 1^3+2^3+0^3 = 9$. Then you can calculate $e_1=3, e_2=2, e_3=0$ and plug them into the formula for $p_3$: $e_1^3 - 3e_1e_2 + 3e_3 = 3^3 - 3(3)(2) + 3(0) = 27 - 18 = 9$. The numbers match perfectly, as they must.

These identities are computational powerhouses. For example, if you need to express $p_6$ in terms of elementary polynomials, you can just mechanically apply these rules step-by-step to build up from $p_1$ to $p_6$. The process is completely determined. It's an algorithm. This algorithmic nature is a cornerstone of a deep result called the ​​Fundamental Theorem of Symmetric Polynomials​​, which guarantees that any symmetric polynomial can be written as a unique combination of elementary ones.

So, we have this beautiful mathematical machinery. But what is it for? One of the most immediate and profound applications is in understanding the roots of polynomials.

Consider a polynomial, say $P(t) = t^n + c_1 t^{n-1} + \dots + c_n$. The Fundamental Theorem of Algebra tells us it has $n$ roots in the complex numbers, let's call them $\lambda_1, \lambda_2, \dots, \lambda_n$. Now, here's the magic: the coefficients $c_k$ of the polynomial are nothing but the elementary symmetric polynomials of its roots (with a sign change): $c_k = (-1)^k e_k(\lambda_1, \dots, \lambda_n)$.

So the polynomial's coefficients encode the "collective interaction" information of its roots. But what about the power sums of the roots, $p_k = \sum \lambda_i^k$? These quantities often have direct physical or mathematical meaning. Newton's identities provide the link. If you know the coefficients of a polynomial, you can use the identities to calculate any power sum of its roots without ever finding the roots themselves!

For example, given the polynomial $P(x) = x^4 - 2x^3 + 3x^2 - 5x + 7$, we can immediately read off the elementary symmetric polynomials of its roots: $e_1=2, e_2=3, e_3=5, e_4=7$. If we want to find the sum of the sixth powers of the roots, $p_6$, we don't need to solve a messy quartic equation. We just turn the crank on Newton's identities and, after a few steps, find that $p_6 = -41$. We have learned something deep about the roots without ever seeing them.

Sometimes, the structure of the identities simplifies beautifully. Consider a system where the first four elementary symmetric polynomials are zero, but $e_5 = -2$. A quick application of Newton's identities shows that $p_1=p_2=p_3=p_4=0$, but $p_5=-10$. For higher powers, the identities reduce to a simple recurrence: $p_k = -2 p_{k-5}$. This allows one to find $p_{15}$ almost instantly as $-40$. The abstract identities reveal hidden patterns.

The connection becomes even more astonishing when we step into the realm of linear algebra. Every square matrix $A$ has a set of characteristic numbers associated with it, its ​​eigenvalues​​ $\lambda_1, \dots, \lambda_n$. These numbers are, in a sense, the "soul" of the matrix, describing how it stretches and rotates space.

The elementary symmetric polynomials of these eigenvalues, $e_k(\lambda_1, \dots, \lambda_n)$, appear as the coefficients of the matrix's ​​characteristic polynomial​​, $p(\lambda) = \det(\lambda I - A)$. Finding eigenvalues can be extremely difficult.

But what about the power sums of the eigenvalues, $p_k = \sum \lambda_i^k$? Miraculously, this quantity is equal to something very easy to compute: the ​​trace​​ (the sum of the diagonal elements) of the matrix $A$ raised to the $k$-th power. $p_k = \sum_{i=1}^n \lambda_i^k = \operatorname{tr}(A^k)$

Think about what this means. You can compute $\operatorname{tr}(A)$, $\operatorname{tr}(A^2)$, $\operatorname{tr}(A^3)$, etc., just by matrix multiplication and addition—no roots required. These are the power sums of the hidden eigenvalues. Now, using Newton's identities, you can take these experimentally accessible trace values and convert them into the elementary symmetric polynomials. And those give you the coefficients of the characteristic polynomial!.

So, just by looking at the traces of the powers of a matrix, we can reconstruct its fundamental DNA—the characteristic polynomial—without ever solving for the eigenvalues. This is a powerful and unexpected bridge between the brute-force computation of matrix powers and the subtle, intrinsic properties of a linear transformation.

The relationship between $p_k$ and $e_k$ is so fundamental that it can be captured in even more elegant forms. The entire system of Newton's identities can be solved, using methods like Cramer's rule, to give a direct formula for any $e_k$ in terms of the power sums. This formula takes the shape of a ​​determinant​​.

Now that we have acquainted ourselves with the intricate dance between power sums and elementary symmetric polynomials, you might be asking, "What good is it?" It is a fair question. Are these identities merely a piece of algebraic trivia, a clever puzzle for mathematicians? Or do they represent something deeper, a pattern that nature itself finds appealing? The wonderful answer is that this relationship is not just a curiosity; it is a fundamental pattern that echoes through an astonishing variety of scientific fields, from the most abstract realms of pure mathematics to the tangible theories of modern physics. It is one of those master keys that unlocks seemingly unrelated doors, and by following it, we can begin to see the beautiful, unified structure of the mathematical world.

Let us begin our journey with the most immediate application: solving equations. Suppose you have a system of variables, but instead of knowing them individually, you only know their collective properties—their sum, their sum of pairwise products, and so on. These are, of course, the elementary symmetric polynomials. If you then need to calculate a different collective property, say, the sum of the squares or cubes of these variables, you are essentially trying to find a power sum. Newton's identities provide the exact machine for this task. They are the dictionary that translates between these two natural languages for describing a system of numbers, allowing us to compute quantities like $x^2+y^2+z^2$ or $x^3+y^3+z^3$ directly from the basic symmetric information given.

This street goes both ways. Imagine you are a detective trying to identify an unknown polynomial. The roots of the polynomial are the culprits, but they are in hiding. You have some clues, however: you know the sum of the roots, the sum of their squares, and their product. This is a mixture of power sums and elementary symmetric polynomials. Using Newton's identities, you can convert all your clues into the language of elementary symmetric polynomials. And what are those? By Vieta's formulas, they are precisely the coefficients of the polynomial you're looking for! You have just reconstructed the polynomial's identity without ever having to find the individual roots. This idea is not just a puzzle; it forms the basis of many algorithms in computational algebra.

This connection becomes even more profound when we step into the world of linear algebra. Every square matrix has a set of characteristic numbers associated with it, its eigenvalues. They are, in a sense, the most important numbers describing the matrix's behavior. The sum of the eigenvalues is the trace of the matrix, $\operatorname{tr}(A)$, and their product is the determinant, $\det(A)$. But what about other combinations? A beautiful fact is that the trace of a matrix raised to a power, $\operatorname{tr}(A^k)$, is exactly the $k$-th power sum of its eigenvalues, $\sum \lambda_i^k$.

Suddenly, our abstract power sums have a concrete physical and geometric meaning. This allows us to use traces, which are easy to compute, to uncover the coefficients of the characteristic polynomial of a matrix, whose roots are the all-important eigenvalues. This link is so fundamental that it forms the bedrock of invariant theory. If you are looking for properties of a matrix that do not change when you change your coordinate system (a process called conjugation), you are looking for functions $P(A)$ such that $P(gAg^{-1}) = P(A)$. It turns out that any polynomial property with this invariance—no matter how complicated—can be expressed as a polynomial in these simple traces: $\operatorname{tr}(A), \operatorname{tr}(A^2), \dots, \operatorname{tr}(A^n)$. These power sums are the fundamental building blocks of all polynomial invariants. This principle can even be extended from polynomials to all continuous invariant functions, showing that these traces form a complete set of "coordinates" for any property that depends only on eigenvalues.

The same pattern continues to appear, like a familiar refrain in a grand symphony, as we move to even more abstract fields.

In Galois theory, which studies the symmetries of the roots of polynomials, the "trace" of an element in a field extension is defined as the sum of its images under all the symmetries of the extension. For an element $\alpha^k$, this trace is precisely the $k$-th power sum of the conjugates of $\alpha$—which are themselves the roots of its minimal polynomial. Once again, Newton's identities connect the coefficients of this fundamental polynomial to the traces of its powers.

In the study of special functions, we often encounter sequences of orthogonal polynomials, such as the Chebyshev polynomials, which are critical in approximation theory and the study of differential equations. If we need to know the sum of the fourth powers of the roots of the fourth Chebyshev polynomial, we don't need to solve a complicated quartic equation. We can simply write down the polynomial's coefficients and use Newton's sums as a mechanical recipe to find the answer.

Perhaps the most breathtaking appearance of this pattern is in the world of geometry and topology. To classify complex vector bundles—geometric objects that attach a vector space to every point on a manifold—topologists use characteristic classes, such as Chern classes $c_k(E)$. Through a clever device known as the splitting principle, these Chern classes behave exactly like elementary symmetric polynomials. When we consider a complex bundle as a real bundle, we can describe it using different classes, the Pontryagin classes $p_k(E_{\mathbb{R}})$. How are these two descriptions related? It turns out that the first Pontryagin class is given by the formula $p_1(E_{\mathbb{R}}) = 2c_2(E) - c_1(E)^2$. This structure is deeply connected to the algebraic power sum $p_2 = e_1^2 - 2e_2$, corresponding to its negative. The very structure of space and geometry is written in the same language as the roots of a simple polynomial!

Finally, let's look to the frontiers of theoretical physics. In trying to modify Einstein's theory of general relativity, some physicists explore "bimetric theories" that use two different metrics to describe spacetime. To construct the interaction between these two metrics, they need to build scalar quantities that are independent of the coordinate system. The fundamental building blocks are the elementary symmetric polynomials, $e_k$, of the eigenvalues of a matrix that mixes the two metrics. And how are these terms constructed in practice? They are built from the power sums—the traces of the powers of that matrix, $[K], [K^2], [K^3], \dots$—using Newton's identities as the blueprint. The formulas developed by Newton in the 17th century are being used today to write down potential new laws of gravity.

From solving simple equations to describing the fabric of spacetime, the elegant dance between power sums and symmetric polynomials is a universal one. It shows us that the insights we gain in one small corner of the scientific world can have unexpected and profound implications everywhere else. This is the great joy of discovery: finding these threads of unity that tie the whole magnificent tapestry together.