Spectrum of an Operator

In the world of finite-dimensional linear algebra, eigenvalues offer a complete picture of a matrix's scaling properties. But what happens when we transition to the infinite-dimensional spaces of functional analysis, which are essential for describing systems in physics and engineering? The simple notion of an eigenvalue is no longer sufficient to capture the complex behavior of operators. This article tackles this gap by introducing the powerful and more general concept of the ​​spectrum of an operator​​.

First, we will deconstruct the spectrum, moving beyond eigenvalues to explore its three fundamental components—the point, continuous, and residual spectra. We will uncover how to determine the spectrum and learn about powerful shortcuts like the Spectral Mapping Theorem. Subsequently, we will reveal why this abstract concept is indispensable, demonstrating how it serves as the very language of quantum mechanics, linking operator properties to measurable physical quantities like energy and position, and answering fundamental questions about stability and existence in physical systems.

If you've ever encountered linear algebra, you've probably met the idea of an ​​eigenvalue​​. For a square matrix $A$ in a finite-dimensional space, you look for special vectors that are only scaled by the matrix, not changed in direction. These are the eigenvectors, and the scaling factors, the numbers $\lambda$, are the eigenvalues. Finding them involves solving the equation $(A - \lambda I)v = 0$ for a non-zero vector $v$. This is equivalent to finding $\lambda$ such that the matrix $A - \lambda I$ is "singular"—that is, its determinant is zero and it's not invertible. For matrices, the story pretty much ends there: the set of eigenvalues is the spectrum.

But when we step into the vast, sprawling landscapes of infinite-dimensional spaces—like the space of all continuous functions on an interval, or the space of sound waves—things get much more interesting. An operator $T$ (the infinite-dimensional cousin of a matrix) can fail to have a nice inverse in more subtle ways than just having a zero pop up in a determinant. The concept of "not being invertible" splinters into a fascinating variety of possibilities. This richer collection of "problematic" numbers $\lambda$ is called the ​​spectrum​​ of the operator, denoted $\sigma(T)$.

Formally, the spectrum $\sigma(T)$ is the set of all complex numbers $\lambda$ for which the operator $T - \lambda I$ is not invertible in the strongest sense: it fails to be a one-to-one, onto mapping with a stable, bounded inverse. To truly understand an operator, we can't just look at its eigenvalues; we must explore its entire spectrum. It’s like trying to understand a person not just by their single biggest trait, but by the full range of their personality.

Why can an operator $T - \lambda I$ fail to have a good inverse? It turns out there are three fundamental ways this can happen, and they partition the spectrum into three disjoint sets. Let's meet the cast.

First, there's the most familiar character: the ​​point spectrum​​, $\sigma_p(T)$. This is the set of "true" eigenvalues. For these values of $\lambda$, the operator $T - \lambda I$ fails to be injective; it maps multiple distinct inputs to the same output. In particular, it maps some non-zero vector (an eigenvector) to the zero vector. Finding these is often a direct algebraic exercise. For instance, consider an operator $T$ on the space of continuous functions on $[0,1]$ defined by $(Tf)(x) = f(1) - f(x)$. To find its eigenvalues, we solve $Tf = \lambda f$, which becomes $f(1) - f(x) = \lambda f(x)$. A little detective work reveals that this equation only has non-trivial solutions if $\lambda=0$ (for constant functions) or if $\lambda=-1$ (for functions that are zero at $x=1$). So, the point spectrum is precisely the set $\{-1, 0\}$.

Next up is the ​​continuous spectrum​​, $\sigma_c(T)$. This is where things get truly "infinite-dimensional." For a $\lambda$ in the continuous spectrum, the operator $T - \lambda I$ is injective (no eigenvectors!), and its range is "almost" the whole space (it's a dense subset), but it's not quite onto. More critically, its inverse exists but is unbounded. This means that you can find a sequence of vectors whose outputs under $T - \lambda I$ get closer and closer to zero, even while the inputs stay a fixed size. These are sometimes called "approximate eigenvectors." A beautiful example arises in a model of a one-dimensional crystal lattice, represented by the operator $(Tx)_n = x_{n-1} + 2x_n + x_{n+1}$ on the space of infinite sequences $\ell^2(\mathbb{Z})$. This operator, a discrete version of the second derivative, has no eigenvalues at all! However, using the powerful tool of the Fourier transform, we can see that its spectrum is the entire interval $[0,4]$. Since it has no eigenvalues, this entire interval is its continuous spectrum. In physics, such continuous spectra correspond to "bands" of allowed energies for an electron moving through a crystal, rather than discrete energy levels of an isolated atom.

Finally, we have the ​​residual spectrum​​, $\sigma_r(T)$. This is the third and sometimes most peculiar possibility. Here, $T - \lambda I$ is injective (again, no eigenvectors), but its range is "small"—it's not even a dense subset of the whole space. This means there's a whole portion of the space that you can't even get close to by applying the operator. While this category is crucial for a complete theory, it often doesn't appear for the most common types of operators in physics, the self-adjoint operators. For them, the residual spectrum is always empty.

Abstract definitions are one thing, but intuition thrives on concrete examples. The most intuitive and fundamental class of operators are ​​multiplication operators​​. Imagine an operator $M_f$ whose only job is to take a function $g(x)$ and multiply it by a fixed function $f(x)$, resulting in a new function $(M_f g)(x) = f(x)g(x)$. When would the shifted operator $M_f - \lambda I$ fail to be invertible?

The operator $M_f - \lambda I$ is just multiplication by the function $f(x) - \lambda$. To invert this, you would need to multiply by $1/(f(x) - \lambda)$. But what if, for some point $x_0$, we have $f(x_0) - \lambda = 0$? Then the would-be inverse blows up to infinity at $x_0$, and it's no longer a well-behaved function in our space. This leads to a beautifully simple conclusion: $\lambda$ is in the spectrum of $M_f$ if and only if $\lambda$ is a value that the function $f(x)$ actually takes on. In other words, the spectrum of a multiplication operator is simply the ​​range​​ (or image) of the multiplying function.

For example, if we consider the operator on continuous functions on $[0,1]$ that multiplies by the function $f(t) = t^2 - t$, the spectrum is simply the set of all values that $t^2 - t$ can produce for $t \in [0,1]$. A quick check with calculus shows this function's minimum is $-\frac{1}{4}$ and its maximum is $0$. So, the spectrum $\sigma(M_f)$ is the closed interval $[-\frac{1}{4}, 0]$. This is a wonderful result! The abstract concept of a spectrum boils down to finding the range of a simple function. This principle is at the heart of quantum mechanics, where the position operator $X$, which multiplies a wavefunction $\psi(x)$ by $x$, has a spectrum equal to the range of possible positions.

One of the most elegant aspects of spectral theory is its internal consistency. If you know the spectrum of an operator $T$, you can often figure out the spectrum of a new operator built from $T$—like $T^2$, $3T$, or $(I-T)^{-1}$—without redoing all the hard work. This is the magic of the ​​spectral mapping theorem​​.

The simplest version involves a simple shift. What is the spectrum of $T+cI$, where $c$ is a constant? The operator $(T+cI) - (\lambda+c)I$ is just $T-\lambda I$. So, the shifted operator fails to be invertible for the value $\lambda+c$ precisely when the original operator fails for $\lambda$. This means the new spectrum is just the old spectrum shifted by $c$ in the complex plane: $\sigma(T+cI) = \sigma(T) + c$. So if you have an operator whose spectrum is, say, a shape in the complex plane, adding $3-i$ to the operator simply slides that entire shape 3 units to the right and 1 unit down.

This idea extends far beyond simple shifts. For any polynomial $p(z)$, the spectrum of the operator $p(T)$ is just the set of values you get by applying the polynomial to every point in the spectrum of $T$. That is, $\sigma(p(T)) = p(\sigma(T)) = \{p(\lambda) \mid \lambda \in \sigma(T)\}$. This is a fantastically powerful tool. Suppose we know the spectrum of the multiplication-by-$x$ operator on $L^2([0,1])$ is the interval $[0,1]$. What is the spectrum of the operator $A = i(M^2 - M)$? We just take the polynomial $p(z) = i(z^2 - z)$ and apply it to every point in $[0,1]$. As we saw before, the function $z^2-z$ maps $[0,1]$ to $[-\frac{1}{4}, 0]$. Multiplying by $i$ rotates this segment onto the imaginary axis. The result is that $\sigma(A)$ is the line segment from $-i/4$ to $0$. An operation that looked complicated becomes a simple exercise in mapping a set of numbers. This incredible theorem even works for more complex functions, like rational functions, making it a cornerstone of the field.

The structure of the spectrum can also reveal deep symmetries of the operator itself.

A key operation is taking the ​​adjoint​​ of an operator, $T^*$, which is the infinite-dimensional analogue of the conjugate transpose of a matrix. The spectrum of the adjoint is related to the original spectrum in a very simple way: it's the complex conjugate of it. That is, $\sigma(T^*) = \overline{\sigma(T)} = \{\bar{\lambda} \mid \lambda \in \sigma(T)\}$. Geometrically, this is just a reflection across the real axis in the complex plane. If an operator has a spectrum that's the line segment from $0$ to $i$, its adjoint will have a spectrum that's the line segment from $0$ to $-i$. This has a profound consequence. If an operator is ​​self-adjoint​​ ($T = T^*$), its spectrum must be equal to its own complex conjugate. The only numbers that are their own conjugates are real numbers. Therefore, the spectrum of any self-adjoint operator must lie entirely on the real line. This is why observables in quantum mechanics—like position, momentum, and energy—are represented by self-adjoint operators: their possible measurement outcomes (their spectra) must be real numbers.

Finally, a particularly elegant story unfolds for a class of operators known as ​​compact operators​​. These are operators on infinite-dimensional spaces that are, in a sense, "almost finite-dimensional." They squash infinite bounded sets into sets that are nearly finite. This "squashing" property has a dramatic effect on their spectrum. For a compact operator on an infinite-dimensional space, the spectrum is remarkably tame: it is a countable (or finite) set of points, and these points can only pile up at a single location: zero. A set like the unit disk $|z| \leq 1$ is too "big" and "dense" to be the spectrum of a compact operator. A set like $\{0, 1, 1+i, 1+2i, \dots\}$ is impossible because it is unbounded. But a set like $\{0\} \cup \{1/n \mid n=1, 2, 3, \dots\}$ is a perfect candidate. This beautiful structure theorem tells us that compact operators, despite living in infinite dimensions, have spectra that behave almost as nicely as eigenvalues of a matrix, with the crucial addition of the accumulation point at zero, a ghostly reminder of the infinite space they inhabit.

From its basic definition to its tripartite nature, and through the magic of spectral mapping and the beautiful structures of special cases, the spectrum of an operator provides a deep and detailed portrait of its behavior, unifying algebra, analysis, and physics in a single, powerful concept.

Now that we have acquainted ourselves with the machinery of an operator's spectrum, you might be tempted to ask: What is it all for? Is this intricate collection of numbers, this "spectrum," merely a curio for the mathematical cabinet, a set of abstract properties to be cataloged and admired? The answer is a resounding no! The spectrum of an operator is one of the most powerful and insightful concepts in modern science. It is a bridge, a Rosetta Stone, that connects the abstract, high-level algebra of operators to the tangible, measurable, and often surprising behavior of the physical world. It reveals the very "character" of a system. Let's take a journey to see how.

One of the most elegant features of spectral theory is its predictive power. If you know the spectrum of a single, fundamental operator, you can often deduce the spectrum of a whole family of more complex operators built from it, without having to redo all the difficult analysis from scratch. This is the magic of the ​​spectral mapping theorem​​, which, in its simplest form, states that if you apply a function $f$ to an operator $T$, the new spectrum is just the set of values of $f$ applied to the old spectrum. It’s a veritable "calculus of properties."

Suppose we have an operator $T$ and we create a new one by simply scaling and shifting it: $S = aT + bI$. It feels intuitive that the properties of $S$ should be simply scaled and shifted versions of the properties of $T$. The spectral mapping theorem confirms this with beautiful precision: the spectrum of $S$ is exactly the set of numbers $a\lambda + b$, where $\lambda$ is in the spectrum of $T$. It's as simple as that.

But we can do much more. Consider a projection operator, $P$. This is the quintessential "switch." It projects a vector onto a subspace. For any vector, it's either "on" (if the vector is in the subspace) or "off" (if the vector is in the orthogonal complement). Its spectrum, fittingly, is the set $\{0, 1\}$. Now, what if we construct a new operator, say $T = 2P + 3I$? We are combining our switch with a constant background. What are the possible measurement outcomes for $T$? The spectral mapping theorem gives the answer instantly. The new spectrum is $\{2(0)+3, 2(1)+3\}$, which is $\{3, 5\}$. The "off" state of the system now has a value of 3, and the "on" state has a value of 5. The algebraic structure transparently maps onto the spectral structure.

This principle reveals even more peculiar behaviors. What about a nilpotent operator, $N$—an operator that becomes the zero operator after being applied to itself a few times ($N^k = 0$)? Such an operator represents a kind of "transient" or "decaying" process. Its spectrum is just the single point $\{0\}$. Now, if we build a polynomial operator like $T = \alpha N^2 + \beta N + \gamma I$, the spectral mapping theorem tells us the spectrum of $T$ is just $\{\gamma\}$. All the complicated, nilpotent parts of the operator have become "spectrally invisible," contributing nothing to the set of outcomes except at zero. The spectrum is determined entirely by the simplest part of the operator, the identity!

When the spectrum is not a discrete set of points but a continuous interval, the same idea holds, but we must be a little more careful. Consider the operator $T$ that corresponds to multiplying a function by its variable, $x$, on the interval $[0,1]$. Its spectrum is, naturally, the entire interval $[0,1]$. If we form a new operator $B = T^2 - 2T$, its spectrum is not just the values of $f(t)=t^2-2t$ at the endpoints $t=0$ and $t=1$. We must apply the function to every point in the spectrum of $T$. We must find the complete range of the function $f(t)$ over the interval $[0,1]$, which turns out to be the interval $[-1, 0]$. The spectrum of the new operator is a continuous interval, but its shape and bounds are determined by the behavior of the function we used to create it.

The connection between operator spectra and the real world becomes most profound and most literal in the realm of quantum mechanics. In the early 20th century, physicists made a startling discovery: physical observables—things we can measure, like position, momentum, and energy—are not represented by simple numbers, but by self-adjoint operators acting on a Hilbert space of "states." In this radical new picture, the spectrum of an operator is no longer an analogy; it is the set of all possible outcomes of a measurement of that physical quantity.

This single idea explains one of the most bizarre features of the quantum world: quantization. Why can an electron in a hydrogen atom only have certain discrete energy levels? Because the Hamiltonian operator (the energy operator) for that system has a discrete spectrum. Why can a free particle have any energy? Because its Hamiltonian has a continuous spectrum.

Let's look at the position operator, $\hat{x}$, for a particle moving along the entire real line. We have a gut feeling that the particle could be found anywhere, so the set of possible position measurements should be the entire real line, $(-\infty, \infty)$. And indeed, the spectrum of $\hat{x}$ is exactly that. But why is the spectrum continuous? It's because there are no "true" eigenvectors for position in the Hilbert space of physically allowable states. An eigenstate of position $x_0$ would have to be a function that is zero everywhere except at the single point $x_0$. Such a creature, the Dirac delta function, is a useful mathematical tool, but its square is infinite, so it cannot represent a physical state (it is not in $L^2(\mathbb{R})$). This failure to find normalizable eigenfunctions is the mathematical signature of a continuous spectrum. The spectrum tells us that while we can find the particle in any small interval, we can never find it in a state of perfectly definite position.

The spectrum doesn't just tell us about static measurements; it governs the dynamics of the system. The evolution of a quantum state in time is described by the unitary operator $U(t) = \exp(-iHt/\hbar)$, where $H$ is the Hamiltonian, or energy operator. If the spectrum of $H$ is the set of allowed energies $\{E_n\}$, then the spectral mapping theorem tells us that the spectrum of $U(t)$ is the set of phase factors $\{\exp(-iE_n t/\hbar)\}$. All these spectral values lie on the unit circle in the complex plane, which is the hallmark of a unitary operator—one that preserves the total probability. The energy spectrum of the system dictates the frequencies of its "internal clock." The allowed energies are the notes, and the time evolution is the music that is played.

Even simplified models of physical systems are beautifully illuminated by spectral theory. Imagine a particle hopping along a one-dimensional crystal lattice. This can be modeled by the bilateral shift operator $B$ on the space of sequences $\ell^2(\mathbb{Z})$. A related operator, $T = B - B^{-1}$, can be seen as a discrete version of a momentum or derivative operator. The spectrum of the basic shift operator $B$ is the unit circle. Applying the spectral mapping theorem with the function $f(z) = z - 1/z$, we find that the spectrum of $T$ is the purely imaginary interval $[-2i, 2i]$. This interval is the "energy band" of our simple crystal, the continuous range of energies available to an electron moving through the lattice.

Beyond specific calculations, the spectrum of an operator reveals deep structural truths about the system it describes. It can answer fundamental questions of existence and stability.

For instance, we know how to take the square root of a positive real number. Can we always find a self-adjoint "square root" $S$ for any self-adjoint operator $T$, such that $S^2 = T$? The spectrum gives us an immediate and decisive "no." A self-adjoint operator $S$ must have a real spectrum. Its square, $T=S^2$, must therefore be a "positive operator," meaning its spectrum can only contain non-negative numbers. This is because for any eigenvector $v$ of $S$ with eigenvalue $\lambda$, we have $Tv = S^2v = S(\lambda v) = \lambda^2 v$. The eigenvalues of $T$ are the squares of the eigenvalues of $S$, and are therefore non-negative. So, if we are handed a self-adjoint operator $T$ that has even a single strictly negative value in its spectrum, we know, without doing any further calculation, that it cannot have a self-adjoint square root. The spectrum acts as a fundamental certificate of positivity.

Perhaps one of the most profound applications lies in understanding how systems respond to perturbations. Consider a free particle moving in space. Its energy operator $A = -\frac{d^2}{dx^2}$ has a continuous spectrum $[0, \infty)$, representing that the particle can have any non-negative kinetic energy. Now, what happens if we introduce a potential, like the Coulomb attraction that binds an electron to a proton? We are adding a new operator $K$ to our Hamiltonian: $B = A+K$. If this potential $K$ is "localized" or "fades away at infinity" (which mathematically corresponds to $K$ being a compact operator), then Weyl's theorem on the essential spectrum gives a spectacular result: the continuous part of the spectrum does not change! It remains $[0, \infty)$.

What does this mean? It means the perturbation cannot change the physics of particles that are far away from the potential's influence; they can still fly by with any kinetic energy. These are the scattering states. However, the perturbation can do something else. It can create new, isolated eigenvalues, typically below the essential spectrum. These are the bound states! For the hydrogen atom, these are the famous discrete energy levels. Spectral theory thus provides the perfect mathematical framework to distinguish between a particle that is bound to a system and one that is just passing by.

From a simple set of numbers, we have journeyed to the heart of quantum reality. The spectrum is far more than a mathematical abstraction. It is the DNA of an operator, encoding its fundamental properties, its relationship to others, and the physical story of the system it describes. By learning to read this spectrum, we see the beautiful unity between the world of abstract operators and the concrete, measurable phenomena of nature.