The Matrix Exponential and Linear Flows
~5 min read · Tier toggle changes depth (T1 plain · T2 standard · T3 rigour)
You should be comfortable with the definition of a linear operator, the concept of an eigenvalue problem, and the basic properties of inner products in finite-dimensional vector spaces.
Why this matters
The matrix exponential is the bridge between static linear algebra and dynamic physical systems.
In classical mechanics and quantum theory, the evolution of a state is governed by a linear differential equation. Solving these equations directly is often impossible or cumbersome. The matrix exponential provides a closed-form solution that encapsulates the entire time evolution in a single operator. It allows us to treat the time dependence as a geometric transformation of the state space, preserving the linear structure and enabling the use of spectral decomposition to analyze complex dynamics.
The Idea in Plain Language
Imagine a system where the rate of change of a quantity is always proportional to the quantity itself. In scalar calculus, this leads to exponential growth or decay. In linear algebra, the "quantity" is a vector, and the "proportionality constant" is a matrix. The matrix exponential generalizes the scalar exponential function to operators. It tells us how to "rotate" or "scale" a vector in a continuous manner over time. If the matrix represents a rotation generator, the exponential gives the actual rotation. If it represents a damping process, the exponential gives the decay. Crucially, this operation is smooth and always invertible, \(\exp(tA)^{-1} = \exp(-tA)\), so the flow can be run backwards mathematically, even for a damping generator whose backward evolution is never seen in nature.
Definition and Convergence
The matrix exponential is defined via the power series expansion, mirroring the scalar case. For a bounded linear operator \(A\) on a finite-dimensional space \(V\), the series is given by:
This series converges absolutely for all \(t \in \mathbb{R}\) because the norm of the terms decreases factorially. The resulting operator \(U(t) = \exp(tA)\) is differentiable with respect to \(t\), and its derivative satisfies the differential equation \(\dot{U}(t) = A U(t)\) with the initial condition \(U(0) = I\). This uniqueness ensures that the exponential is the only solution to this operator-valued initial value problem.
Solving Linear Flows
Consider the initial value problem for a vector \(x(t)\):
The solution is obtained by applying the matrix exponential to the initial state:
To verify this, we differentiate with respect to \(t\):
Since \(x(0) = \exp(0) x_{0} = I x_{0} = x_{0}\), the initial condition is satisfied. The family of operators \(\{U(t)\}_{t \in \mathbb{R}}\) forms a one-parameter group, meaning \(U(t+s) = U(t)U(s)\) and \(U(-t) = U(t)^{-1}\). This group property reflects the time-translation invariance of the underlying linear dynamics.
Unitarity and Self-Adjointness
In quantum mechanics, the time evolution operator must be unitary to preserve the norm of the state vector (probability conservation). A key result links this physical requirement to the algebraic properties of the generator \(A\). The operator \(U(t) = \exp(tA)\) is unitary for all \(t\) if and only if \(A\) is skew-Hermitian (or anti-self-adjoint):
If \(A\) is skew-Hermitian, then \(A^{\dagger} = -A\). The inverse of \(U(t)\) is \(U(-t) = \exp(-tA)\). We check unitarity:
Since \(A\) commutes with itself, the exponentials combine: \(\exp(-tA) \exp(tA) = \exp(0) = I\). Thus, \(U(t)^{\dagger} U(t) = I\), confirming unitarity. This result is foundational for the Schrödinger equation, where the Hamiltonian is Hermitian, and the generator is \(-iH/\hbar\), which is skew-Hermitian.
The derivation
Matrix Exponential and One-Parameter Flows
Let \(A\) be a bounded linear operator on a finite-dimensional (or, more generally, a Hilbert) space \(V\). The operator series \(\exp(tA)=\sum_{n=0}^{\infty}\frac{t^{n}A^{n}}{n!}\) converges in operator norm for every \(t\in\mathbb{R}\) and defines the unique differentiable operator-valued function \(U(t)\) satisfying \(\dot U=AU\), \(U(0)=I\). Consequently \(x(t)=U(t)x_{0}\) is the unique solution of the initial-value problem \(\dot x=Ax\), \(x(0)=x_{0}\), the family \(\{U(t)\}_{t\in\mathbb{R}}\) is a one-parameter group, and \(U(t)\) is unitary for all \(t\) if and only if the generator is skew-adjoint, \(A^{\dagger}=-A\) — equivalently \(A=-\tfrac{i}{\hbar}H\) with \(H\) self-adjoint, in which case \(H\) is the physical generator of the flow.
A common pitfall is assuming that \(\exp(A+B) = \exp(A)\exp(B)\). This equality holds only if \(A\) and \(B\) commute, i.e., \(AB = BA\). If they do not commute, the Baker-Campbell-Hausdorff formula introduces additional commutator terms, making the exponential non-abelian. In physical systems with non-commuting generators (e.g., position and momentum), this distinction is critical for correct time-ordered evolution. C
Worked example
Problem. Consider a two-dimensional system governed by \(\dot{x} = A x\) where \(A = \begin{pmatrix} 0 & -1 \\ 1 & 0 \end{pmatrix}\). Find the solution \(x(t)\) for \(x(0) = \begin{pmatrix} 1 \\ 0 \end{pmatrix}\).
- Identify that \(A\) is skew-symmetric, hence \(A^{\dagger} = -A\), so \(U(t)\) will be orthogonal (real unitary).
- Compute \(A^2 = \begin{pmatrix} 0 & -1 \\ 1 & 0 \end{pmatrix} \begin{pmatrix} 0 & -1 \\ 1 & 0 \end{pmatrix} = \begin{pmatrix} -1 & 0 \\ 0 & -1 \end{pmatrix} = -I\).
- Use the series expansion: \(\exp(tA) = I + tA + \frac{t^2 A^2}{2!} + \frac{t^3 A^3}{3!} + \dots = I + tA - \frac{t^2}{2!}I - \frac{t^3}{3!}A + \dots\).
- Group even and odd powers: \(\exp(tA) = \left( \sum_{k=0}^{\infty} \frac{(-1)^k t^{2k}}{(2k)!} \right) I + \left( \sum_{k=0}^{\infty} \frac{(-1)^k t^{2k+1}}{(2k+1)!} \right) A\).
- Recognize the cosine and sine series: \(\exp(tA) = \cos(t) I + \sin(t) A\).
- Substitute \(A\): \(\exp(tA) = \begin{pmatrix} \cos t & -\sin t \\ \sin t & \cos t \end{pmatrix}\).
- Apply to \(x_0\): \(x(t) = \begin{pmatrix} \cos t & -\sin t \\ \sin t & \cos t \end{pmatrix} \begin{pmatrix} 1 \\ 0 \end{pmatrix} = \begin{pmatrix} \cos t \\ \sin t \end{pmatrix}\).
Units check. The matrix \(A\) has units of \(\mathrm{rad/s}\) (angular velocity). The argument \(tA\) is dimensionless (time \(\times\) frequency), which is correct for the exponential function. The resulting vector components are dimensionless, consistent with a normalized state vector.
Three quick tests
- If \(A\) is a real symmetric matrix, is \(\exp(tA)\) orthogonal?
Answer
No. If \(A\) is symmetric, \(A^{\dagger} = A\). Then \(U(t)^{\dagger} U(t) = \exp(tA) \exp(tA) = \exp(2tA) \neq I\) unless \(A=0\). Symmetric generators lead to exponential growth/decay, not rotations. - What is the inverse of \(U(t) = \exp(tA)\)?
Answer
The inverse is \(U(-t) = \exp(-tA)\). This follows from the group property \(U(t)U(-t) = \exp(tA)\exp(-tA) = \exp(0) = I\). - Does \(\exp(A+B) = \exp(A)\exp(B)\) always hold?
Answer
No. It holds if and only if \(AB = BA\). If they do not commute, the product includes commutator corrections via the Baker-Campbell-Hausdorff formula.
Problem set for PU-104 →
Each derivation in this unit ends with five graded problems with full solutions.
Lecture notes drafted with a local model and checked line by line by the editor; every number in the worked example was recomputed.