Motivation
For a scalar equation , integration gives an explicit exponential. Matrix equations are subtler: matrices at different times need not commute, so the scalar-looking formula is not generally a solution. Nevertheless, the determinant of a fundamental matrix satisfies a scalar equation involving only the trace. This is Liouville's formula.
When the coefficient matrix is periodic, the effect of one full period is captured by a single matrix. Repeated periods correspond to its powers. Floquet theory makes that relationship precise and gives a stability test, while also revealing a distinction between a complex exponential representation and a real one. The periodic and logarithm hypotheses will be stated explicitly; an average trace by itself does not classify stability.
Definitions
Definition 4.1.1
(Fundamental matrix and monodromy)
Let be continuous. The normalized fundamental matrix solves
Its columns are the solutions from the standard basis, and is invertible for every . Indeed, a nonzero vector in its kernel would give a solution that is zero at time ; uniqueness backwards would force its initial vector to be zero.
If for some , the matrix is the monodromy matrix for the initial time zero. Its complex eigenvalues, counted with algebraic multiplicity, are the Floquet multipliers. A multiplier is a property of the one-period map; it is not an eigenvalue of at an individual time.
Continuity gives existence and uniqueness for every finite time for this linear equation. The fundamental matrix can also be considered over by complexifying the real system. Stability of the zero solution below refers to the usual forward-time notion: sufficiently small real initial conditions stay small for all ; asymptotic stability also requires their solutions to tend to zero.
Determinants and one-period dynamics
Theorem 4.1.2
(Liouville's determinant formula)
For continuous and the normalized fundamental matrix ,
More generally, any fundamental matrix satisfies . For a periodic system, the product of all Floquet multipliers is therefore
Theorem 4.1.3
(Floquet representation over the complex numbers)
If is continuous and -periodic, there are a constant complex matrix and a complex invertible -periodic matrix such that
The representation need not be unique. Even for real , a representation with both and real and of period need not exist.
Theorem 4.1.4
(Forward stability from multipliers)
For the real periodic system, the zero solution is exponentially stable if and only if every multiplier has modulus strictly less than one. It is stable, without necessarily being asymptotically stable, if and only if every multiplier has modulus at most one and every multiplier on the unit circle is semisimple: its complex Jordan blocks all have size one. A multiplier of modulus greater than one, or a nontrivial Jordan block on the unit circle, makes zero unstable.
These are statements about the whole monodromy matrix. A determinant smaller than one says that the product of multiplier moduli is smaller than one; it does not prevent one multiplier from exceeding one.
From continuous time to a matrix power
Proof
(The trace controls the determinant)
Since is invertible, Jacobi's determinant identity gives
The last equality uses cyclicity of trace. The initial determinant is one, so solving this scalar linear equation gives the formula. The general initial-time form follows the same way. For a scalar second-order equation , the associated first-order matrix has trace , giving Abel's Wronskian formula .
Proof
(Periodicity and the complex matrix logarithm)
The matrix satisfies and . Uniqueness for the matrix initial-value problem gives
Liouville's formula makes invertible. Every invertible complex matrix has a complex logarithm: put it in Jordan form, choose a complex logarithm of each nonzero eigenvalue , and for a nilpotent Jordan part use the finite polynomial
on a block of size . The nilpotent exponential and logarithm series terminate and are inverse as formal polynomials, so exponentiating the result recovers the block. Assemble the blocks and take to be a logarithm of divided by .
Set . It is invertible and
because commutes with . Choosing different logarithm branches can produce different and . Nothing in this construction makes or real.
Proof
(Why multipliers control forward stability)
For , with integer and , repeated use of periodicity yields
Both and are uniformly bounded on the compact interval . Thus boundedness, decay or unbounded growth of the powers of determines the corresponding behavior between period endpoints as well as at them.
Complex Jordan form shows that exponentially when every eigenvalue has modulus below one: the polynomial factors from a Jordan block are dominated by a slightly slower exponential. Conversely, an eigenvalue outside the unit disk makes the powers unbounded. An eigenvalue on the unit circle contributes bounded powers precisely when all its Jordan blocks have size one; a larger block contributes a growing polynomial in . The spectral conclusions hold for real initial data as well: if a complex vector has unbounded images, at least one of its real and imaginary parts does.
Uniform boundedness of is equivalent to stability for this linear system. Exponential decay of gives exponential stability; if all solutions tend to zero, then , forcing all multipliers inside the unit disk. This proves the classification.
Proof X-ray
What the determinant can and cannot tell us
The proof reduces continuous time to integer powers through ; this identity uses only periodicity of and uniqueness. Compactness of makes the reduction two-sided: the uniform bounds on and let boundedness, decay or growth pass from to and back, so nothing can happen between period endpoints that the powers do not see.
The non-obvious step is on the unit circle, where the modulus alone does not decide. A Jordan block of size for a multiplier contributes entries of size , and semisimplicity is exactly what removes the polynomial factor. Reality of is used once, to pass from complex to real initial data. Liouville's formula enters only to make invertible: it controls the product of the multipliers, while the proof needs each multiplier separately, so a shrinking determinant cannot replace it.
Worked examples
Worked example 4.1.5
(A scalar periodic coefficient)
Consider , where . Integration gives
The zero solution is exponentially stable if , stable but not asymptotically stable if , and unstable if . The oscillatory factor can cause transient variation, but it is bounded above and below. Here the trace is the entire scalar coefficient, so its period integral determines the only multiplier; that special simplification does not extend to general systems.
Worked example 4.1.6
(A shrinking determinant can hide an expanding direction)
Take the constant system with and . Its monodromy is
The determinant agrees with , but the first coordinate doubles every period. Zero is unstable. Volume contraction alone is not contraction in every direction.
Worked example 4.1.7
(A unit multiplier with a Jordan block)
For and ,
Both multipliers equal one, but the Jordan block is not semisimple. For initial vector , the first coordinate is , which is unbounded for every nonzero . This shows why the unit-circle condition needs semisimplicity.
Two tempting shortcuts that fail
Counterexample mode
A real periodic system can lack a real T-periodic Floquet exponent
Let , ,
Set . Because , this is a real continuous one-periodic matrix. Its normalized fundamental matrix is , so
There is no real matrix with . Such a commutes with its exponential ; because has distinct diagonal entries, any real matrix commuting with it is real diagonal. The exponential of a real diagonal matrix has positive diagonal entries, a contradiction. A real -periodic Floquet representation normalized by has and would imply , giving a real logarithm of . The complex representation still exists. Positive determinant alone does not guarantee a real logarithm. In this example does have a real logarithm, so the same construction gives a real representation with period .
Counterexample mode
Without commutation, the solution need not be the exponential of the integral
Let and . On set ; on set . This is continuous, and each scalar coefficient has integral one on its interval. Thus
The matrices are different. The first interval acts before the second, so order matters. The coefficient can be extended continuously with period two because it vanishes at both endpoints; periodicity alone does not make the matrices commute.
Common mistakes
Common mistake
(Reading stability from the instantaneous eigenvalues)
For the -periodic matrix
one has and for every , so every has the eigenvalues , with negative real part. Yet direct substitution shows that is a solution, and . Thus is a multiplier of modulus greater than one, and zero is unstable. Liouville's formula is consistent with this: , so the other multiplier is .
Common mistake
(Treating Liouville's formula as a formula for )
Liouville's formula is a statement about only. In the second counterexample above, and both have determinant , as the zero trace requires, yet they are different matrices. The trace computation is scalar, whereas the time-ordered action of a matrix system retains directional information.
Common mistake
(Checking moduli but not Jordan blocks)
In the Jordan-block example both multipliers equal , so a test based only on would declare zero stable, while the solution through grows like . On the unit circle, semisimplicity must be checked as well.
A Floquet matrix logarithm is not unique, and a real logarithm need not exist with the original period. None of these statements alone covers nonlinear stability of a periodic orbit; that requires a separate perturbation and return-map analysis.
Summary
Liouville's formula turns a fundamental determinant into an integral of the trace. Periodicity gives , and a complex logarithm of supplies a Floquet representation. The full spectrum and unit-circle Jordan structure of determine forward linear stability. The determinant and instantaneous eigenvalues are less informative than the monodromy matrix itself.
Exercises
Exercise 4.1.8
(For , find the multiplier over and classify zero.)
Foundation
Solution(2)
- 1
Integrate the coefficient
, and .
- 2
Read off the multiplier
, so zero is exponentially stable.
Exercise 4.1.9
(For , where is continuous, find in terms of for two solutions. Does the formula require to be periodic?)
Foundation
Solution(2)
- 1
Write the first-order system
For the matrix is , whose trace is , and is the determinant of a fundamental matrix.
- 2
Apply Liouville's formula
. Only continuity of the coefficients is used; does not even appear, and periodicity is not needed.
Exercise 4.1.10
(Let . Classify the zero solution of a real periodic linear system with this monodromy, and justify the answer from its powers.)
Core
Solution(3)
- 1
Compute the powers
, which is uniformly bounded in but does not tend to zero.
- 2
Transfer to continuous time
Key step
with bounded on , so is bounded for , and the solution starting at does not tend to zero.
- 3
Classify
Zero is stable but not asymptotically stable; the unit multiplier is semisimple. This monodromy is realizable: for period one take .
Exercise 4.1.11
(A periodic system has period and trace . Find . Explain why the determinant alone does not determine stability.)
Core
Solution(3)
- 1
Apply Liouville's formula
, the product of the multipliers.
- 2
Exhibit an unstable system with this trace
Key step
has trace and gives , with a multiplier larger than one, so zero is unstable.
- 3
Exhibit a stable one
has the same trace and , so zero is exponentially stable. The determinant is the same in both cases, so it cannot decide.