Exercise 4
Physics-Aware Machine Learning
Dr.-Ing. Annika Junker
Winter term 2025/2026
Task 4.1: Third-order ODE
You are given the following third-order linear differential equation with an external
input u(t):
...
x (t) + 6ẍ(t) + 11ẋ(t) + 6x(t) = u(t).
(a) Define suitable state variables x1 (t), x2 (t), x3 (t) so that the equation can be
rewritten as a system of three first-order differential equations.
(b) Write down the resulting system of three coupled first-order equations.
(c) Express the system in matrix form:
ẋ(t) = Ax(t) + bu(t)
⊤
with the state vector x(t) = x1 (t), x2 (t), x3 (t) .
(d) Compute the eigenvalues of matrix A. What do they tell you about the
stability of the system?
Solution:
(a) State variables: x1 (t) = x(t), x2 (t) = ẋ(t), x3 (t) = ẍ(t).
ẋ1 (t) = x2 (t)
(b) First-order system: ẋ2 (t) = x3 (t)
ẋ3 (t) = −6x1 (t) − 11x2 (t) − 6x3 (t) + u(t).
(c) Matrix form: ẋ(t) = Ax(t) + bu(t)
0 1 0 0
with A = 0 0 1 , b = 0.
−6 −11 −6 1
(d) Calculating the characteristic polynomial:
det(A − λI) = λ3 + 6λ2 + 11λ + 6 = 0
(λ + 1)(λ + 2)(λ + 3) = 0
yields λ1 = −1, λ2 = −2, λ3 = −3. All eigenvalues have negative real
parts which means that the system is asymptotically stable.
1
Exercise 4 Physics-Aware Machine Learning
Task 4.2: Characterizing dynamical systems
List and briefly explain at least 6 criteria that can be used to characterize a dynam-
ical system. Your goal is to develop a structured understanding of how systems
differ and what properties are important for modeling, simulation, and control.
You may consider mathematical, physical, and technical aspects.
Solution:
Criterion Description Example values
Time invariance whether system parameters time-invariant /
change over time time-variant
Time behavior whether the system evolves continuous-time /
continuously or in discrete steps discrete-time
Linearity whether the system obeys the linear / nonlinear
principle of superposition
Stability whether the system returns to stable / unstable /
equilibrium or diverges over time marginally stable
Autonomy whether the system’s dynamics autonomous /
depend explicitly on time or non-autonomous
external inputs.
Determinism whether the system’s future deterministic /
behavior is fully determined by stochastic
its current state and inputs
Input-output number of inputs and outputs SISO / MIMO
structure
Controllability whether the system can be fully / partially / not
driven to any state using controllable
suitable inputs
Observability whether the internal state can be fully / partially / not
inferred from outputs observable
Physical the type of energy or signals mechanical / electrical /
domain involved thermal / hydraulic etc.
Task 4.3: State-space representation of a damped mass on
a spring
Consider a mass m attached to a linear spring with spring constant c and subject
to an external acting force F and a linear damping force with damping coefficient
d. According to Newton’s second law, the differential equation is given by
mẍ = −cx − dẋ + F. (1)
A sensor measures the position x(t) of the mass.
(a) Define suitable state variables so that the second-order differential equation
can be rewritten as a system of two first-order differential equations.
2
Exercise 4 Physics-Aware Machine Learning
(b) Write down the resulting state-space system in matrix form.
Solution:
(a) State variables: x1 (t) = x(t), x2 (t) = ẋ(t).
(
ẋ1 (t) = x2 (t)
(b) State-space system:
ẋ2 (t) = − mc x1 (t) − md x2 (t) + 1
m
F (t),
or in matrix form ẋ(t) = Ax(t) + bu,
0 1 0
with A = c d , b = 1 , u = F (t).
−m −m m
Task 4.4: State-space representation of a pendulum
The simple pendulum is a nonlinear dynamical system consisting of a mass sus-
pended from a fixed point by a massless rod, with its motion governed by a non-
linear ordinary differential equation
g d
φ̈ = − sin φ − 2 φ̇,
l ml
with the angle φ, the gravitational acceleration g and the length l and the damping
d of the pendulum.
(a) Define suitable state variables so that the second-order differential equation
can be rewritten as a system of two first-order differential equations.
(b) Write down the resulting state-space system in matrix form.
Solution:
(a) State variables: x1 (t) = φ(t), x2 (t) = φ̇(t).
(
ẋ1 (t) = x2 (t)
(b) State-space system:
ẋ2 (t) = − gl sin x1 (t) − d
x (t),
ml2 2