Numerical Evaluation of Dynamic Response

Ajay Singhal
September 26, 2024
 Numerical Evaluation of Dynamic Response
▪ Analytical methods solve the differential equation of motion exactly and provide the solution u(t) as a
continuous function
▪ Analytical solution of the equation of motion is usually not possible for general loadings or for nonlinear
systems

▪ Numerical methods discretize the problem into time steps to obtain discrete approximation of u(t)

▪ Usually numerical methods are the only way to find solutions to complex loading and nonlinear problems

▪ Approaches
▪ Piecewise Exact Method
▪ Central Difference Method
▪ Newmark’s Method

Singhal 2
 Piecewise Exact Method
▪ Highly efficient numerical procedure for linear elastic systems
▪ Interpolate the excitation over each time interval. For short time intervals, linear interpolation is satisfactory
∆𝑝
▪ 𝑝 𝜏 = 𝑝𝑖 + ∆𝑡 𝑖 𝜏
𝑖

∆𝑝
▪ Over the time step, we need to solve: 𝑚𝑢ሷ + 𝑐𝑢ሶ + 𝑘𝑢 = 𝑝𝑖 + ∆𝑡 𝑖 𝜏
𝑖

∆𝑝
▪ For simplicity, let’s neglect damping: 𝑚𝑢ሷ + 𝑘𝑢 = 𝑝𝑖 + ∆𝑡 𝑖 𝜏
𝑖

▪ Using solutions we have already studied and “Initial Conditions” of 𝑢𝑖 and 𝑢ሶ 𝑖 :

𝑢ሶ 𝑝𝑖 ∆𝑝𝑖 𝜏 sin 𝜔𝑛 𝜏
▪ 𝑢 𝜏 = 𝑢𝑖 cos 𝜔𝑛 𝜏 + 𝜔 𝑖 sin 𝜔𝑛 𝜏 + 1 − cos 𝜔𝑛 𝜏 + −
𝑛 𝑘 𝑘 ∆𝑡𝑖 𝜔𝑛 ∆𝑡𝑖
ሶ
𝑢(𝜏) 𝑢ሶ 𝑝𝑖 ∆𝑝𝑖 1
▪ Differentiating gives: = −𝑢𝑖 sin 𝜔𝑛 𝜏 + 𝜔 𝑖 cos 𝜔𝑛 𝜏 + sin 𝜔𝑛 𝜏 + 1 − cos 𝜔𝑛 𝜏
𝜔𝑛 𝑛 𝑘 𝑘 𝜔𝑛 ∆𝑡𝑖

▪ The displacement and velocity can be evaluated at 𝜏 = ∆𝑡𝑖 to give displacement 𝑢𝑖+1 and velocity 𝑢ሶ 𝑖+1
𝑢ሶ 𝑝𝑖 ∆𝑝𝑖 1
▪ 𝑢𝑖+1 = 𝑢𝑖 cos 𝜔𝑛 ∆𝑡𝑖 + 𝜔 𝑖 sin 𝜔𝑛 ∆𝑡𝑖 + 1 − cos 𝜔𝑛 ∆𝑡𝑖 + 𝜔𝑛 ∆𝑡𝑖 − sin 𝜔𝑛 ∆𝑡𝑖
𝑛 𝑘 𝑘 𝜔𝑛 ∆𝑡𝑖
𝑢ሶ 𝑖+1 𝑢ሶ 𝑝𝑖 ∆𝑝𝑖 1
▪ = −𝑢𝑖 sin 𝜔𝑛 ∆𝑡𝑖 + 𝜔 𝑖 cos 𝜔𝑛 ∆𝑡𝑖 + sin 𝜔𝑛 ∆𝑡𝑖 + 1 − cos 𝜔𝑛 ∆𝑡𝑖
𝜔𝑛 𝑛 𝑘 𝑘 𝜔𝑛 ∆𝑡𝑖

Singhal 3
 Piecewise Exact Method (Cont’d)
▪ The equations for displacement and velocity can
be rewritten as:
▪ 𝑢𝑖+1 = 𝐴𝑢𝑖 + 𝐵𝑢ሶ 𝑖 + 𝐶𝑝𝑖 + 𝐷𝑝𝑖+1
▪ 𝑢ሶ 𝑖+1 = 𝐴′ 𝑢𝑖 + 𝐵′ 𝑢ሶ 𝑖 + 𝐶 ′ 𝑝𝑖 + 𝐷 ′ 𝑝𝑖+1
where 𝐴, 𝐵,… 𝐶 ′ , 𝐷 ′ are given in Table 5.2.1
for damping as well as no damping cases.

▪ Notes
▪ Accuracy of the method depends on how
accurate the linear approximation of p(t) is
▪ Exact solution to approximate p(t)
▪ Good for linear SDOF problems only

Singhal 4
 Central Difference Method
▪ Based on finite difference approximation of velocity 𝑢ሶ and acceleration 𝑢ሷ u
▪ For constant ∆𝑡, we can approximate 𝑢ሶ and 𝑢ሷ by: ui
𝑢𝑖+1 −𝑢𝑖−1 ui-1
▪ 𝑢ሶ 𝑖 = 2 ∆𝑡
𝑑𝑢 𝑑𝑢
∆𝑡 ∆𝑡
− 𝑢𝑖+1 −𝑢𝑖 𝑢𝑖 −𝑢𝑖−1
𝑑2 𝑢 𝑑𝑡 𝑡 +∆𝑡 𝑑𝑡 𝑡 −∆𝑡 − 𝑢𝑖+1 −2𝑢𝑖 + 𝑢𝑖−1
𝑖 2 𝑖 2
▪ 𝑢ሷ 𝑖 = = = ∆𝑡 ∆𝑡
= ti-1 ti ti+1 t
𝑑𝑡 2 𝑡𝑖 ∆𝑡 ∆𝑡 ∆𝑡 2

▪ Equation of motion at time t = ti: 𝑚𝑢ሷ 𝑖 + 𝑐𝑢ሶ 𝑖 + 𝑘𝑢𝑖 = 𝑝𝑖

𝑢𝑖+1 −2𝑢𝑖 + 𝑢𝑖−1 𝑢𝑖+1 −𝑢𝑖−1

▪ Substituting for 𝑢ሶ 𝑖 and 𝑢ሷ 𝑖 from above, we get: 𝑚 +𝑐 + 𝑘𝑢𝑖 = 𝑝𝑖
∆𝑡 2 2 ∆𝑡

෠ 𝑖+1 = 𝑝Ƹ 𝑖 where
▪ Gathering terms, we can write: 𝑘𝑢
𝑚 𝑐
▪ 𝑘෠ = + 2∆𝑡 : Effective Stiffness
∆𝑡 2

𝑚 𝑐 2𝑚
▪ 𝑝Ƹ 𝑖 = 𝑝𝑖 − ∆𝑡 2
− 2∆𝑡 𝑢𝑖−1 − 𝑘 − ∆𝑡 2
𝑢𝑖 : Effective Load

Singhal 5
 Central Difference Method (Cont’d)
▪ Need displacement at previous two steps (𝑢𝑖−1 and 𝑢𝑖 ): Two step method

▪ Two step methods are difficult to start since at first we know only the initial conditions: 𝑢0 and 𝑢ሶ 0

▪ Need an expression for 𝑢−1 to start. Can use finite difference expressions to find 𝑢−1
𝑢1 −𝑢−1 𝑢1 −2𝑢0 + 𝑢−1
▪ 𝑢ሶ 0 = and 𝑢ሷ 0 =
2 ∆𝑡 ∆𝑡 2

▪ Solving for 𝑢1 from the first equation: 𝑢1 = 2∆𝑡 𝑢ሶ 0 + 𝑢−1 . Substituting in the second equation above,
∆𝑡 2
gives: 𝑢−1 = 𝑢0 − ∆𝑡 𝑢ሶ 0 + 𝑢ሷ 0
2
𝑝0 −𝑐𝑢ሶ 0 −𝑘𝑢0
▪ 𝑢ሷ 0 can be found from the equation of motion at t = 0: 𝑢ሷ 0 =
𝑚

▪ Now we have everything to start stepping through

Singhal 6
 Central Difference Method – Steps
𝑝0 −𝑐𝑢ሶ 0 −𝑘𝑢0
1. Calculate 𝑢ሷ 0 =
𝑚

∆𝑡 2
2. Calculate 𝑢−1 = 𝑢0 − ∆𝑡 𝑢ሶ 0 + 𝑢ሷ 0
2
𝑚 𝑐
3. Calculate 𝑘෠ = +
∆𝑡 2 2∆𝑡

𝑚 𝑐
4. Calculate 𝑎 = −
∆𝑡 2 2∆𝑡

2𝑚
5. Calculate 𝑏 = 𝑘 −
∆𝑡 2

6. 𝑝Ƹ 𝑖 = 𝑝𝑖 − 𝑎𝑢𝑖−1 − 𝑏𝑢𝑖
𝑝ො𝑖
7. 𝑢𝑖+1 = ෠
𝑘

𝑢𝑖+1 −𝑢𝑖−1 𝑢𝑖+1 −2𝑢𝑖 + 𝑢𝑖−1

8. 𝑢ሶ 𝑖 = and 𝑢ሷ 𝑖 =
2 ∆𝑡 ∆𝑡 2

9. Replace i by i+1
Singhal 7
 Central Difference Method – Notes
▪ The central difference method will “blow-up” giving meaningless results in the presence of numerical round-
off if the time step chosen is not short enough

∆𝑡 1
▪ For numerical stability <
𝑇𝑛 𝜋

▪ Thus Central Difference Method is called a “Conditionally Stable” method. Never really a constraint for
SDOF systems because a much smaller time step should be chosen to obtain results that are accurate.
∆𝑡
Typically ≤ 0.1 to define the response adequately.
𝑇𝑛

Singhal 8
 Newmark’s Method
▪ A family of time-stepping methods based on the following relations:

▪ 𝑢ሶ 𝑖+1 = 𝑢ሶ 𝑖 + 1 − 𝛾 ∆𝑡 𝑢ሷ 𝑖 + 𝛾∆𝑡 𝑢ሷ 𝑖+1

1 2 2
▪ 𝑢𝑖+1 = 𝑢𝑖 + ∆𝑡 𝑢ሶ 𝑖 + − 𝛽 ∆𝑡 𝑢ሷ 𝑖 + 𝛽 ∆𝑡 𝑢ሷ 𝑖+1
2

▪ Parameters 𝛽and 𝛾 define

▪ The variation of acceleration over the time step

▪ Also define the accuracy and stability of the method

▪ The above two equations combined with the equilibrium equation at the end of the time step can be used to
compute 𝑢𝑖+1 , 𝑢ሶ 𝑖+1 , and 𝑢ሷ 𝑖+1

▪ Iteration required in this formulation since 𝑢ሷ 𝑖+1 appears on the right-hand side of the above equations

Singhal 9
 Newmark’s Method – Incremental Form
▪ Incremental equation of motion can be written as: 𝑚 ∆𝑢ሷ 𝑖 + 𝑐 ∆𝑢ሶ 𝑖 + 𝑘 ∆𝑢𝑖 = ∆𝑝𝑖 .…………..(1)
where ∆𝑢ሷ 𝑖 = 𝑢ሷ 𝑖+1 − 𝑢ሷ 𝑖 , ∆𝑢ሶ 𝑖 = 𝑢ሶ 𝑖+1 − 𝑢ሶ 𝑖 , ∆𝑢𝑖 = 𝑢𝑖+1 − 𝑢𝑖 , and ∆𝑝𝑖 = 𝑝𝑖+1 − 𝑝𝑖
▪ Newmark’s equations can be written in incremental form as:
▪ ∆𝑢ሶ 𝑖 = ∆𝑡 𝑢ሷ 𝑖 + 𝛾∆𝑡 ∆𝑢ሷ 𝑖 ……..…………………………….…………….…………..(2)
∆𝑡 2 2
▪ ∆𝑢𝑖 = ∆𝑡 𝑢ሶ 𝑖 + 𝑢ሷ 𝑖 + 𝛽 ∆𝑡 ∆𝑢ሷ 𝑖 …………….……........…………………..(3)
2
1 1 1
▪ Equation (3) can be rewritten as: ∆𝑢ሷ 𝑖 = ∆𝑢𝑖 − 𝑢ሶ − 𝑢ሷ .………………….……..(4)
𝛽 ∆𝑡 2 𝛽∆𝑡 𝑖 2𝛽 𝑖
𝛾 𝛾 𝛾
▪ Substituting equation (4) into equation (2) gives: ∆𝑢ሶ 𝑖 = ∆𝑢𝑖 − 𝑢ሶ 𝑖 + 1 − 𝑢ሷ 𝑖 ……………...(5)
𝛽∆𝑡 𝛽 2𝛽
෠ 𝑖 = ∆𝑝Ƹ 𝑖
▪ Now substituting equations (4) and (5) into equation of motion given by equation (1), we get 𝑘∆𝑢
where
𝛾 1
▪ 𝑘෠ = 𝑘 + 𝑐+ 𝑚 : Effective Stiffness
𝛽∆𝑡 𝛽 ∆𝑡 2
1 𝛾 1 𝛾
▪ ∆𝑝Ƹ 𝑖 = ∆𝑝𝑖 + 𝑚 + 𝑐 𝑢ሶ 𝑖 + 𝑚 + ∆𝑡 − 1 𝑐 𝑢ሷ 𝑖 : Effective Load
𝛽∆𝑡 𝛽 2𝛽 2𝛽

Singhal 10
 Newmark’s Method – Steps
𝑝0 −𝑐𝑢ሶ 0 −𝑘𝑢0
1. Calculate 𝑢ሷ 0 =
𝑚
𝛾 1
2. Calculate 𝑘෠ = 𝑘 + 𝑐 + 𝑚 : Effective Stiffness
𝛽∆𝑡 𝛽 ∆𝑡 2
1 𝛾
3. Calculate 𝐴 = 𝑚 + 𝑐
𝛽∆𝑡 𝛽
1 𝛾
4. Calculate 𝐵 = 𝑚 + ∆𝑡 −1 𝑐
2𝛽 2𝛽

5. ∆𝑝Ƹ 𝑖 = ∆𝑝𝑖 + 𝐴𝑢ሶ 𝑖 + 𝐵𝑢ሷ 𝑖 : Effective Load

∆𝑝ො𝑖
6. ∆𝑢𝑖 = ෠
𝑘
𝛾 𝛾 𝛾
7. ∆𝑢ሶ 𝑖 = ∆𝑢𝑖 − 𝑢ሶ 𝑖 + 1 − 𝑢ሷ 𝑖
𝛽∆𝑡 𝛽 2𝛽
1 1 1
8. ∆𝑢ሷ 𝑖 = ∆𝑢𝑖 − 𝑢ሶ − 𝑢ሷ
𝛽 ∆𝑡 2 𝛽∆𝑡 𝑖 2𝛽 𝑖

9. 𝑢𝑖+1 = 𝑢𝑖 + ∆𝑢𝑖 , 𝑢ሶ 𝑖+1 = 𝑢ሶ 𝑖 + ∆𝑢ሶ 𝑖 , and 𝑢ሷ 𝑖+1 = 𝑢ሷ 𝑖 + ∆𝑢ሷ 𝑖

10. Replace i by i+1

Singhal 11
 Average Acceleration Method
1
▪ Average acceleration 𝑢ሷ 𝜏 = 𝑢ሷ 𝑖+1 + 𝑢ሷ 𝑖
2
𝜏
▪ After integrating and using BC at  = 0, we get 𝑢ሶ 𝜏 = 𝑢ሶ 𝑖 + 𝑢ሷ 𝑖+1 + 𝑢ሷ 𝑖
2

𝜏2
▪ After integration again and using BC at  = 0, we get 𝑢 𝜏 = 𝑢𝑖 + 𝑢ሶ 𝑖 𝜏 + 𝑢ሷ 𝑖+1 + 𝑢ሷ 𝑖
4

From the last two equations, we get

∆𝑡
▪ 𝑢ሶ 𝑖+1 = 𝑢ሶ 𝑖 + 𝑢ሷ 𝑖+1 + 𝑢ሷ 𝑖
2

(∆𝑡)2
▪ 𝑢𝑖+1 = 𝑢𝑖 + 𝑢ሶ 𝑖 ∆𝑡 + 𝑢ሷ 𝑖+1 + 𝑢ሷ 𝑖
4

1 1
Same as Newmark’s equations with β = 4 and 𝛾 = 2

Singhal 12
 Linear Acceleration Method
𝜏
▪ Linear acceleration 𝑢ሷ 𝜏 = 𝑢ሷ 𝑖 + 𝑢ሷ 𝑖+1 − 𝑢ሷ 𝑖
∆𝑡

𝜏2
▪ After integrating and using BC at  = 0, we get 𝑢ሶ 𝜏 = 𝑢ሶ 𝑖 + 𝑢ሷ 𝑖 𝜏 + 𝑢ሷ 𝑖+1 − 𝑢ሷ 𝑖
2∆𝑡

𝜏2 𝜏3
▪ After integration again and using BC at  = 0, we get 𝑢 𝜏 = 𝑢𝑖 + 𝑢ሶ 𝑖 𝜏 + 𝑢ሷ 𝑖 2 + 6∆𝑡
𝑢ሷ 𝑖+1 + 𝑢ሷ 𝑖

From the last two equations, we get

∆𝑡
▪ 𝑢ሶ 𝑖+1 = 𝑢ሶ 𝑖 + 2
𝑢ሷ 𝑖+1 + 𝑢ሷ 𝑖

1 1
▪ 𝑢𝑖+1 = 𝑢𝑖 + 𝑢ሶ 𝑖 ∆𝑡 + (∆𝑡)2 6 𝑢ሷ 𝑖+1 + 3 𝑢ሷ 𝑖

1 1
Same as Newmark’s equations with β = 6 and 𝛾 = 2

Singhal 13
 Numerical Stability
▪ Numerical procedures that result in bounded solutions only if ∆𝑡 is shorter than some stability limit are termed
Conditionally Stable
∆𝑡 1 1
▪ Newmark’s methods are stable if ≤𝜋
𝑇𝑛 2 𝛾−2𝛽

▪ If there is no such limit on ∆𝑡, then the procedure is termed Unconditionally Stable
1
▪ Newmark’s methods are Unconditionally Stable if 2𝛽 ≥ 𝛾 ≥ 2
1 1
▪ Stability of Average Acceleration Method (β = 4 and 𝛾 = 2):
1 1 1 1
Check if 2𝛽 ≥ 𝛾 ≥ 2 ֜ 2 4
≥ 2 ≥ 2 is satisfied, therefore Unconditionally Stable
1 1
▪ Stability of Linear Acceleration Method (β = 6 and 𝛾 = 2):
1 1 1 1 1 1 1
Check if 2𝛽 ≥ 𝛾 ≥ 2 ֜ 2 ≥2≥2 ֜ ≥ 2 ≥ 2 is not satisfied, therefore Conditionally Stable when
6 3

∆𝑡 1 1 1 1 ∆𝑡 3 ∆𝑡
≤ = ֜ ≤ or ≤ 0.551
𝑇𝑛 𝜋 2 𝛾−2𝛽 𝜋 2 1 1 𝑇𝑛 𝜋 𝑇𝑛 Stability condition usually not a problem
2
−2 6
as small ∆𝑡 is needed for accuracy
∆𝑡 1
• Central difference method is stable if ≤
𝑇𝑛 𝜋
Singhal 14
 Computational Error
▪ Error expressed in terms of Amplitude Decay
(AD) and Period Elongation (PE) comparing
numerical solutions to theoretical solution for
free vibration
▪ All three methods (average acceleration, linear
acceleration, and central difference) give very
good results with no amplitude decay
∆𝑡
▪ For 𝑇 less than the stability limit, linear
𝑛
acceleration gives the least period elongation

Singhal 15
Singhal 16
 A= 0.8129 A' = -3.5795 𝑢𝑖+1 = 𝐴𝑢𝑖 + 𝐵 𝑢ሶ 𝑖 + 𝐶𝑝𝑖 + 𝐷𝑝𝑖+1
B= 0.09067 B' = 0.7559
C= 0.01236 C' = -0.1709 𝑢ሶ 𝑖+1 = 𝐴′ 𝑢𝑖 + 𝐵′ 𝑢ሶ 𝑖 + 𝐶 ′ 𝑝𝑖 + 𝐷′ 𝑝𝑖+1
D= 0.006352 D' = 0.1871

Singhal 17