Operator Square Root

The concept of a square root is one of the first abstract ideas we encounter in mathematics. While finding the square root of a positive number is straightforward, the notion becomes far more complex and powerful when we ask the same question of operators—the mathematical rules that describe actions like rotation, stretching, and transformation. How do you find the square root of an action, and what does it even mean? This question reveals that our simple intuition from numbers is insufficient, creating a knowledge gap that requires a more robust framework.

This article provides a comprehensive exploration of the operator square root, guiding you from its fundamental principles to its most profound applications. It will illuminate the necessary mathematical machinery that makes the operator square root a well-defined and unique object. You will learn not only how to conceptualize this root but also why it is an indispensable tool across modern science.

The journey is structured in two parts. First, the "Principles and Mechanisms" chapter will establish the rigorous mathematical foundation, explaining why the concepts of positive and self-adjoint operators are crucial and how the spectral theorem provides the key to unlocking the solution. Then, the "Applications and Interdisciplinary Connections" chapter will reveal how this abstract concept is a master key used to define the size of operators, decompose complex transformations, and even describe the fundamental nature of quantum reality. Our exploration begins by dissecting the rules that govern these powerful mathematical objects.

Let's start our journey with a question so familiar it feels almost trivial: what is the square root of 9? The immediate answer is 3. But of course, $(-3)^2$ is also 9. We have two choices. By convention, when we write $\sqrt{9}$, we mean the positive one, the so-called ​​principal square root​​. This choice seems arbitrary, but it gives us a consistent, single-valued function. What about $\sqrt{-1}$? To answer that, we had to invent a whole new class of numbers, the complex numbers. These simple observations—the need for a principal choice and the existence of numbers without real square roots—are like shadows cast by a much grander structure, one that governs not just numbers, but operators.

An operator is a rule, an action. Think of it as a machine: you put a vector in, and it gives you a (possibly different) vector back. A rotation is an operator. A stretch is an operator. In the language of linear algebra, these are represented by matrices. Squaring an operator, $A^2$, is simply the act of applying the same transformation twice in a row. So, finding the square root of an operator $T$ is to ask: is there another operator, let's call it $S$, such that applying $S$ twice is equivalent to applying $T$ just once? That is, can we find an $S$ such that $S^2 = T$?

Let's imagine the simplest kind of operator: one that just stretches space along its axes. In matrix form, this is a diagonal matrix. For example, consider an operator in a two-level quantum system described by a diagonal matrix, like the one in problem.

What is its square root? Here, our intuition serves us perfectly. The operator stretches the first basis vector by $\frac{3}{4}$ and the second by $\frac{1}{4}$. The operator that does this in two "half-steps" would be one that stretches by $\sqrt{\frac{3}{4}}$ and $\sqrt{\frac{1}{4}}$ respectively.

It's simple, clean, and it works. But this simplicity is deceptive. What if one of the numbers on the diagonal was negative? What if the matrix wasn't diagonal at all?

This brings us to a fundamental wall, a direct parallel to trying to find the real square root of a negative number. Let's consider a physically meaningful class of operators called ​​self-adjoint​​ operators. These are the workhorses of quantum mechanics, representing observable quantities like energy, momentum, and position. They have the special property that their "eigenvalues"—the special scaling factors along their preferred axes—are always real numbers.

Now, suppose we are looking for a self-adjoint square root $S$ for another operator $T$. If such an $S$ exists, so that $S^2 = T$, something wonderful happens. For any vector $v$, the "energy" or "expectation value" associated with the operator $T$ is given by the inner product $\langle Tv, v \rangle$. Let's see what this is:

Because $S$ is self-adjoint, we can move it from one side of the inner product to the other.

The quantity $\|Sv\|^2$ is the squared length of the vector $Sv$. A squared length can never be negative. It's always greater than or equal to zero. This means that if an operator $T$ has a self-adjoint square root, it must be a ​​positive operator​​—an operator for which $\langle Tv, v \rangle \ge 0$ for every single vector $v$.

This is a profound conclusion, as highlighted in problem. If you have a self-adjoint operator with even one negative eigenvalue $\lambda_0 < 0$, it cannot be positive (just check $\langle Tv_0, v_0 \rangle = \lambda_0 \|v_0\|^2 < 0$ for its eigenvector $v_0$). Therefore, it is impossible to find a self-adjoint square root for it. We have hit a wall, but in doing so, we've discovered the correct path forward: the search for square roots is a search within the realm of positive operators.

So, for any ​​positive, self-adjoint operator​​ $A$, can we find a square root? And if so, how many are there? This is where the magic of the ​​spectral theorem​​ comes in. In essence, the spectral theorem tells us that for any self-adjoint operator, we can always find a special set of perpendicular axes (a basis of eigenvectors) where the operator's complicated action simplifies to pure stretching. In this special basis, the operator behaves just like a simple diagonal matrix, with its eigenvalues on the diagonal.

Since our operator $A$ is positive, all its eigenvalues are non-negative. In its special "diagonal" basis, finding its square root is as easy as our first example. We just take the positive square root of every eigenvalue! This procedure gives us a new operator, $S$. This operator $S$ is also positive (its eigenvalues are all positive), it is self-adjoint, and by its very construction, $S^2=A$.

Furthermore, because we insisted on taking the positive square root of each eigenvalue, this resulting operator $S$ is unique. We call it ​​the positive square root of A​​, denoted $\sqrt{A}$.

This single, powerful idea is the key to almost every practical calculation.

• When faced with a non-diagonal matrix like in problem or, the underlying method to find the root is to find a way to diagonalize the matrix (or its blocks) and operate on the eigenvalues.
• When we need to find the square root of an operator defined by its effect on basis vectors, as in problem, we simply take the square root of its eigenvalues: an operator that scales the $n$-th basis vector by $\lambda_n = \frac{1}{(|n|+1)^2}$ has a square root that scales it by $\sqrt{\lambda_n} = \frac{1}{|n|+1}$.
• This "functional calculus" is remarkably consistent. For instance, just as $(\sqrt{x})^{-1} = \sqrt{x^{-1}}$ for a positive number $x$, the same holds for operators: the inverse of the square root is the square root of the inverse, $(\sqrt{T})^{-1} = \sqrt{T^{-1}}$, a property explored in.

The concept truly shows its power when we move from finite lists of numbers (vectors in $\mathbb{C}^n$) to functions. In spaces like $L^2$, our "vectors" are functions, such as a sound wave or a quantum wavefunction.

Consider a ​​multiplication operator​​, as in. This operator $T$ takes a function $f(x)$ and returns a new function, $m(x)f(x)$. The operator is "diagonal" in the position basis; at each point $x$, it just multiplies the function's value by the number $m(x)$. If $m(x)$ is a non-negative function, say $m(x)=(x^2+1)^2$, then $T$ is a positive operator. What is its square root? It is, just as you'd guess, the operator that multiplies the function by $\sqrt{m(x)} = x^2+1$. It's beautifully simple.

Now for the grand finale. Let's look at one of the cornerstones of quantum physics: the kinetic energy operator, $A = -\frac{d^2}{dx^2}$. This operator is certainly not a simple multiplication. But the ​​Fourier transform​​ gives us a pair of magic glasses. It allows us to see any function not as a profile in space, $f(x)$, but as a recipe of frequencies (or momenta), $\hat{f}(k)$. When we look at our operator $A$ through these glasses, it is revealed to be a simple multiplication operator! The seemingly complex action of taking the second derivative in position space becomes the simple action of multiplying by $k^2$ in momentum space.

As we saw in problem, finding the square root is now trivial. We just take the square root of the multiplication function: $\sqrt{k^2} = |k|$. Transforming back from momentum space, we find the operator $\sqrt{A}$, which corresponds to multiplying a function's Fourier transform by $|k|$. This operator represents the magnitude of the momentum! We have used the abstract machinery of operator square roots to uncover a deep physical connection between kinetic energy and momentum.

This powerful idea of a function of an operator, defined via the spectral theorem, extends to almost any sensible function, not just the square root. And it has a wonderful consistency. For example, if a bounded operator $B$ commutes with $A$ ($AB=BA$), it will also commute with $\sqrt{A}$ ($B\sqrt{A} = \sqrt{A}B$). As explored in, multiplication operators commute with each other, but they generally don't commute with operators that "mix" points, like differentiation or integration.

From a simple choice about the sign of $\sqrt{9}$, we have journeyed to a principle that unifies finite matrices and infinite-dimensional function spaces, revealing the hidden structure of the fundamental operators of quantum physics. This is the beauty of mathematics: to find a simple, powerful idea that suddenly illuminates a vast landscape of seemingly disconnected problems.

Having established the firm mathematical ground on which the operator square root stands, we might be tempted to view it as a beautiful but esoteric piece of abstract machinery. Nothing could be further from the truth. Like a master key, this single concept unlocks profound insights across an astonishing range of scientific disciplines. The journey from its definition to its application is a tour through the very heart of modern physics, mathematics, and even data science. It is a story that begins with the simple question of "how big" an operator is and ends with us describing the geometry of quantum reality and the nature of motion itself.

Let's begin with a wonderfully intuitive idea. For any complex number $z$, its magnitude or "size" is given by $|z| = \sqrt{\bar{z}z}$. This feels right; we multiply the number by its conjugate to get a non-negative real number (its squared magnitude), and then we take the familiar square root. Can we perform a similar trick for an operator $A$?

The analogue of the complex conjugate for an operator is its adjoint, $A^*$. Following our intuition, we can form the combination $A^*A$. A remarkable fact is that this new operator is always positive and self-adjoint, meaning its "eigen-directions" are orthogonal and its "eigen-stretches" are all non-negative real numbers. This is precisely the condition we need to uniquely define its positive square root. And so, we arrive at a magnificent definition for the "size" or ​​modulus​​ of an operator $A$:

This isn't just a formal definition. For a concrete operator, like a $2 \times 2$ matrix acting on vectors in a plane, we can explicitly compute this modulus matrix. The result is a new operator that encapsulates the pure "stretching" action of the original, stripped of any rotation or reflection. This modulus operator, born from the square root, gives us a rigorous and computable way to talk about the intrinsic magnitude of a linear transformation.

The modulus $|A|$ is far more than just a measure of size; it is the secret to dissecting the operator $A$ itself. One of the most powerful tools in all of linear algebra is the ​​Singular Value Decomposition (SVD)​​. The SVD tells us that any linear transformation, no matter how complicated, can be broken down into three simple steps: a rotation, followed by a pure scaling along perpendicular axes, followed by another rotation.

Where does the operator square root fit in? The "pure scaling" part of the SVD is governed by the singular values, $\{s_n\}$, which are the scaling factors along each principal axis. And these singular values are nothing more than the eigenvalues of the modulus operator, $|A|$! The operator $|A| = \sqrt{A^*A}$ is the operator that performs the scaling. Its spectral decomposition is a magnificent expression of this fact:

This formula, which falls directly out of the SVD framework, tells us that the action of $|A|$ is to take a vector $x$, find its components along the special input directions $\{e_n\}$, and simply scale each component by the corresponding singular value $s_n$. The square root has allowed us to isolate the very heart of the operator's stretching action.

So far, we have used the square root to analyze operators. But can we turn the tables and use an operator to redefine the space it acts upon? Astonishingly, yes. Any positive operator $T$ allows us to define a new way of measuring lengths and angles in our Hilbert space.

Using the unique positive square root $\sqrt{T}$, we can define a new sesquilinear form, a kind of "weighted" inner product:

Imagine the operator $\sqrt{T}$ as a kind of lens that warps the space before we measure the angle between vectors. This new geometry is perfectly valid, but it may have some strange features. For instance, it might be possible for a non-zero vector $x$ to have zero "length" in this new world, i.e., $\langle x, x \rangle_T = 0$. When does this happen? It happens if and only if $\sqrt{T}x = 0$. A little more work reveals that the set of these "null" vectors is precisely the kernel of the original operator $T$ itself. This is a deep and beautiful connection: the algebraic properties of an operator (its kernel) are directly mapped to the geometric properties (the null directions) of the space it helps create.

Nowhere does the operator square root play a more starring role than in the theater of quantum mechanics. Here, it is not merely a useful tool; it is part of the fundamental language used to describe reality.

​​Observables and Energy​​

In the quantum world, physical quantities like position, momentum, and energy are represented by self-adjoint operators. The possible measured values of the quantity are the eigenvalues of its operator. The most important operator is the Hamiltonian, $\hat{H}$, which represents the total energy of a system. For a stable system, energy must be non-negative, which means the Hamiltonian is a positive operator.

This immediately begs the question: can we define a new physical observable whose "square" is energy? Using our trusted tool, we can define the operator $\hat{S} = \sqrt{\hat{H}}$. Because $\hat{H}$ is positive and self-adjoint, its square root $\hat{S}$ is also positive and self-adjoint. This guarantees that $\hat{S}$ is a legitimate physical observable whose measured values will be real and non-negative—specifically, they will be the square roots of the system's energy levels. This might represent a new physical property, derived from energy, whose existence is guaranteed by the mathematics of operator square roots.

​​Measuring Closeness in the Quantum Realm​​

How do you tell if two quantum states are similar? This is a critical question in quantum computing and information theory. The answer is given by a quantity called ​​Bures fidelity​​, a measure of distinguishability between two quantum states (represented by density operators $\rho$ and $\sigma$). Its definition is a breathtaking application of the operator square root:

Look at this marvelous construction! We "sandwich" one state $\sigma$ by the square root of the other state $\rho$. We then take the square root of that entire new operator. Finally, we take its trace and square the result. This intricate dance of square roots is not an arbitrary formula; it is a profoundly geometric measure that correctly captures the notion of "distance" in the strange, curved space of quantum states.

​​A Surprising Symmetry​​

The quantum world is full of surprising symmetries. Consider two self-adjoint operators, $A$ and $B$, that anticommute, meaning $AB = -BA$. (Such relationships appear, for instance, in Dirac's theory of the electron). If we look at the operator $(A+B)^2$, the anticommutation relation causes the cross-terms to cancel out perfectly: $(A+B)^2 = A^2 + AB + BA + B^2 = A^2 + B^2$. This implies that the positive square root of $(A+B)^2$ is simply $\sqrt{A^2+B^2}$. This is a kind of Pythagorean theorem for operators, a hint of the deep algebraic structures, like Clifford algebras, that underpin fundamental physics.

Our exploration has so far focused on square roots of operators representing static properties. But the concept is powerful enough to handle operators that describe processes unfolding in time—dynamics and randomness.

​​Square Roots of Randomness​​

Many random processes in nature, like the jiggling of a particle in a fluid (Brownian motion), can be described using integral operators. For example, the operator $K$ with kernel $k(s,t) = \min(s,t) - st$ is intimately related to a "Brownian bridge"—a random path that starts and ends at the same point. This operator encodes the covariance of the process. We can, of course, take its square root, $K^{1/2}$. What does that mean? The "total energy" of this square root operator, measured by its Hilbert-Schmidt norm, turns out to be directly related to the total variance of the random process. In a beautiful mathematical twist, one can show that $\|K^{1/2}\|_{HS}^2 = \text{Tr}(K)$, the trace of the original operator, which is easily calculated. The square root connects the geometry of the operator space to the statistics of the random process it describes.

​​Square Roots of Dynamics​​

Finally, let's consider the operators that generate motion. The self-adjoint generator of translation (shifting) is the momentum operator, $P = -i\frac{d}{dx}$. Its square, $P^2 = -\frac{d^2}{dx^2}$, is the positive-definite kinetic energy operator, which, in a different context, can be seen as the generator of the diffusion or heat-flow process. This reveals a profound relationship: the square of the generator of translation is the generator of diffusion.

This flips the initial question: what is the square root of the generator of diffusion, $A = -\frac{d^2}{dx^2}$? As established earlier using the Fourier transform, its unique positive square root $\sqrt{A}$ is the operator that multiplies a function's momentum-space representation by $|k|$. This operator, known as the fractional Laplacian, does not generate simple shifting. Instead, it generates a more complex stochastic process called a Lévy flight, which is crucial in modeling anomalous diffusion and long-range interactions. This insight is the gateway to a whole universe of "fractional" dynamics, where one can take a non-integer power of a generator to create entirely new physical processes.