Anonymous
×
Create a new article
Write your page title here:
We currently have 105 articles on MOR Wiki. Type your article name above or click on one of the titles below and start writing!



MOR Wiki

Difference between revisions of "Moment-matching method"

m
 
(9 intermediate revisions by 4 users not shown)
Line 1: Line 1:
 
[[Category:method]]
 
[[Category:method]]
[[Category:DAE order unspecified]]
+
[[Category:Differential_algebraic_system]]
 
[[Category:linear]]
 
[[Category:linear]]
 
[[Category:first differential order]]
 
[[Category:first differential order]]
Line 8: Line 8:
 
==Description==
 
==Description==
 
The moment-matching methods are also called the ''Krylov'' subspace methods<ref name="freund03"/>, as well as
 
The moment-matching methods are also called the ''Krylov'' subspace methods<ref name="freund03"/>, as well as
''Padé'' approximation methods<ref name="feldmann95"/>. They are [[Projection based MOR]]methods. These methods are applicable for non-parametric linear time invariant systems, often the descriptor systems, e.g.
+
''Padé'' approximation methods<ref name="feldmann95"/>. They belong to the [[Projection based MOR]] methods. These methods are applicable to non-parametric linear time invariant systems, often descriptor systems, e.g.
   
 
:<math>E \dot{x}(t)=A x(t)+B u(t),</math>
<math>
 
  +
:<math>y(t)=Cx(t).</math>
E \frac{dx(t)}{dt}=A x(t)+B u(t), \quad
 
y(t)=Cx(t), \quad \quad (1)
 
</math>
 
   
They are very efficient in many engineering applications, circuit simulation, Microelectromechanical systems (MEMS) simulation, etc..
+
They are very efficient in many engineering applications, such as circuit simulation, Microelectromechanical systems (MEMS) simulation, etc..
   
 
The basic steps are as follows. First, the transfer function
 
The basic steps are as follows. First, the transfer function
   
<math>H(s)=Y(s)/U(s)=C(sE-A)^{-1}B</math>
+
:<math>H(s)=Y(s)/U(s)=C(sE-A)^{-1}B</math>
   
 
is expanded into a power series at an expansion point <math>s_0\in\mathbb{C}\cup \infty</math>.
 
is expanded into a power series at an expansion point <math>s_0\in\mathbb{C}\cup \infty</math>.
Line 25: Line 23:
 
Let <math>s=s_0+\sigma</math>, then, within the convergence radius of the series, we have
 
Let <math>s=s_0+\sigma</math>, then, within the convergence radius of the series, we have
   
<math>H(s_0 + \sigma)= C[(s_{0}+\sigma){E}-A]^{-1}B</math>
+
:<math>H(s_0 + \sigma)= C[(s_{0}+\sigma){E}-A]^{-1}B</math>
   
<math>=C[\sigma { E}+(s_{0}{ E}-{ A})]^{-1}B</math>
+
::<math>=C[\sigma { E}+(s_{0}{ E}-{ A})]^{-1}B</math>
   
<math>=C[{ I}+\sigma(s_0{ E}-{ A})^{-1}E]^{-1}[(s_0{ E}-{ A})]^{-1}B</math>
+
::<math>=C[{ I}+\sigma(s_0{ E}-{ A})^{-1}E]^{-1}(s_0{ E}-{ A})^{-1}B</math>
   
<math>=C[{ I}-\sigma(s_0{ E}- A )^{-1}E+\sigma^2[(s_0{ E}-{ A})^{-1}E]^{2}+\ldots]
+
::<math>=C[{ I}-\sigma(s_0{ E}- A )^{-1}E+\sigma^2[(s_0{ E}-{ A})^{-1}E]^{2}+\ldots]
s_0{E}-{ A})^{-1}B</math>
+
(s_0{E}-{ A})^{-1}B</math>
   
<math>=\sum \limits^\infty_{i=0}\underbrace{C[-(s_0{ E}-{A})^{-1}E]^i(s_0{ E}-{ A})^{-1}B}_{:= m_i(s_0)} \, \sigma^i,</math>
+
::<math>=\sum \limits^\infty_{i=0}\underbrace{C[-(s_0{ E}-{A})^{-1}E]^i(s_0{ E}-{ A})^{-1}B}_{:= m_i(s_0)} \, \sigma^i,</math>
   
 
where <math>m_i(s_0)</math> are called the moments of the transfer function about <math>s_0</math> for <math>i=0,1,2,\ldots</math>.
 
where <math>m_i(s_0)</math> are called the moments of the transfer function about <math>s_0</math> for <math>i=0,1,2,\ldots</math>.
If the expansion point is chosen as zero then the moments simplify to <math>m_i(0)=C(A^{-1}E)^i(-A^{-1}B)</math>.
+
If the expansion point is chosen as zero, then the moments simplify to <math>m_i(0)=C(A^{-1}E)^i(-A^{-1}B)</math>.
For <math>s_0=\infty</math> the moments are also called Markov parameters which can be computed by <math>C(E^{-1}A)^i(E^{-1}B)</math> if E is invertable.
 
   
 
The goal in moment-matching model reduction is the construction of a reduced order
 
The goal in moment-matching model reduction is the construction of a reduced order
Line 48: Line 45:
 
example, using a single expansion point <math>s_0=0</math>, by
 
example, using a single expansion point <math>s_0=0</math>, by
   
<math>\textrm{range}(V)=\textrm{span}\{\tilde B,({ A}^{-1}E)^2 \tilde B, \ldots,({ A}^{-1}E)^{r}{\tilde B}\},</math>
+
:<math>\textrm{range}(V)=\textrm{span}\{\tilde B,({ A}^{-1}E)^2 \tilde B, \ldots,({ A}^{-1}E)^{r}{\tilde B}\}, \quad \quad \quad \quad \quad \quad \quad \quad\quad \quad \quad \ (1) </math>
   
<math>\textrm{range}(W)=\textrm{span}\{C^T, E^T{ A}^{-T}C^T,(E^T{A}^{-T})^2C^T, \ldots
+
:<math>\textrm{range}(W)=\textrm{span}\{C^T, E^T{ A}^{-T}C^T,(E^T{A}^{-T})^2C^T, \ldots
,(E^T{A}^{-T})^{r-1}C^T\},</math>
+
,(E^T{A}^{-T})^{r-1}C^T\}, \quad \quad (2) </math>
   
where <math>\tilde B=-A^{-1}B</math>. The reduced model is in the form as below
+
where <math>\tilde B=-A^{-1}B</math>.
 
<math>W^TE \frac{d{V z}}{t}=W^TA Vz +W^T B u(t), \quad \hat{y}(t)=CVz.</math>
 
   
 
The transfer function <math>\hat H</math> of the reduced model has good approximation properties around <math>s_0</math>, which matches the first <math>2r</math> moments of <math>H(s)</math> at <math>s_0</math>.
 
The transfer function <math>\hat H</math> of the reduced model has good approximation properties around <math>s_0</math>, which matches the first <math>2r</math> moments of <math>H(s)</math> at <math>s_0</math>.
   
Using a set of <math>k</math> distinct expansion points <math>\{s_1,\cdots,s_k\}</math>, the reduced model can be obtained by, e.g.,
+
Using a set of <math>k</math> distinct expansion points <math>\{s_1,\cdots,s_k\}</math>, the reduced model obtained by, e.g.,
 
 
   
<math>\textrm{range}(V)=\textrm{span}\{(A-s_1 {E})^{-1}E\tilde B,\ldots,(A-s_k {E})^{-1}E\tilde B \}</math>,
+
:<math>\textrm{range}(V)=\textrm{span}\{(A-s_1 {E})^{-1}E\tilde B,\ldots,(A-s_k {E})^{-1}E\tilde B \}, \quad \quad \quad \quad \quad \quad \quad \quad (3)</math>
  +
 
:<math>\textrm{range}(W)=\textrm{span}\{E^T(A-s_1 {E})^{-T}C^T,\ldots,E^T(A-s_k {E})^{-T}C^T \},\quad \quad \quad \quad \quad (4) </math>
   
 
matches the first two moments at each <math>s_j</math>, <math>j=1,\ldots,k</math>, see <ref name="grimme97"/>. The reduced model is in the form as below
<math>\textrm{range}(W)=\textrm{span}\{E^T(A-s_1 {E})^{-T}C^T,\ldots,E^T(A-s_k {E})^{-T}C^T \},</math>
 
   
 
:<math>W^TE \frac{d{V z}}{t}=W^TA Vz +W^T B u(t), \quad \hat{y}(t)=CVz.</math>
has order <math>r=k</math> and matches the first two moments at each <math>s_j</math>, <math>j=1,\ldots,k</math>, see <ref name="grimme97"/>.
 
   
It can be seen that the columns of <math>V</math>, <math>W</math> span Krylov subspaces
+
For the case of one expansion point in (1)(2), it can be seen that the columns of <math>V</math>, <math>W</math> span Krylov subspaces
  +
which can easily be computed by Arnoldi or Lanczos methods. The matrices <math>V</math> and <math>W</math> in (3)(4) can be computed with the rational Krylov algorithm in<ref name="grimme97"/> or with the modified Gram-Schmidt process. In these algorithms only a few number of linear systems need to be solved, where matrix-vector multiplications are only used if using iterative solvers, which are simple to implement and the complexity of the resulting
which can easily be computed by Arnoldi or Lanczos methods <ref name="freund03"/><ref name="feldmann95"/>. In
 
 
methods is roughly <math>O(n r^2)</math> for sparse matrices <math>A, E</math>.
these algorithms only matrix-vector multiplications are used which
 
are simple to implement and the complexity of the resulting
 
methods is only <math>O(n r^2)</math>.
 
   
 
==References==
 
==References==
Line 82: Line 77:
 
<ref name="feldmann95">P. Feldmann and R.W. Freund, "<span class="plainlinks">[http://dx.doi.org/10.1109/43.384428 Efficient linear circuit analysis by Pade approximation via the Lanczos process]</span>". IEEE Trans. Comput.-Aided Des. Integr. Circuits Syst., 14:639-649, 1995.</ref>
 
<ref name="feldmann95">P. Feldmann and R.W. Freund, "<span class="plainlinks">[http://dx.doi.org/10.1109/43.384428 Efficient linear circuit analysis by Pade approximation via the Lanczos process]</span>". IEEE Trans. Comput.-Aided Des. Integr. Circuits Syst., 14:639-649, 1995.</ref>
   
<ref name="grimme97">E.J. Grimme, "<span class="plainlinks">[http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.19.9254&rep=rep1&type=pdf Krylov projection methods for model reduction]</span>. PhD thesis, Univ. Illinois, Urbana-Champaign, 1997.
+
<ref name="grimme97">E.J. Grimme, "<span class="plainlinks">[http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.19.9254&rep=rep1&type=pdf Krylov projection methods for model reduction]</span>. PhD thesis, Univ. Illinois, Urbana-Champaign, 1997.</ref>
   
 
</references>
 
</references>

Latest revision as of 14:55, 25 August 2022


Description

The moment-matching methods are also called the Krylov subspace methods[1], as well as Padé approximation methods[2]. They belong to the Projection based MOR methods. These methods are applicable to non-parametric linear time invariant systems, often descriptor systems, e.g.

E \dot{x}(t)=A x(t)+B u(t),
y(t)=Cx(t).

They are very efficient in many engineering applications, such as circuit simulation, Microelectromechanical systems (MEMS) simulation, etc..

The basic steps are as follows. First, the transfer function

H(s)=Y(s)/U(s)=C(sE-A)^{-1}B

is expanded into a power series at an expansion point s_0\in\mathbb{C}\cup \infty.

Let s=s_0+\sigma, then, within the convergence radius of the series, we have

H(s_0 + \sigma)= C[(s_{0}+\sigma){E}-A]^{-1}B
=C[\sigma { E}+(s_{0}{ E}-{ A})]^{-1}B
=C[{ I}+\sigma(s_0{ E}-{ A})^{-1}E]^{-1}(s_0{ E}-{ A})^{-1}B
=C[{ I}-\sigma(s_0{ E}- A )^{-1}E+\sigma^2[(s_0{ E}-{ A})^{-1}E]^{2}+\ldots]
(s_0{E}-{ A})^{-1}B
=\sum \limits^\infty_{i=0}\underbrace{C[-(s_0{ E}-{A})^{-1}E]^i(s_0{ E}-{ A})^{-1}B}_{:= m_i(s_0)} \, \sigma^i,

where m_i(s_0) are called the moments of the transfer function about s_0 for i=0,1,2,\ldots. If the expansion point is chosen as zero, then the moments simplify to m_i(0)=C(A^{-1}E)^i(-A^{-1}B).

The goal in moment-matching model reduction is the construction of a reduced order system where some moments \hat m_i of the associated transfer function \hat H match some moments of the original transfer function H.

The matrices V and W for model order reduction can be computed from the vectors which are associated with the moments, for example, using a single expansion point s_0=0, by

\textrm{range}(V)=\textrm{span}\{\tilde B,({ A}^{-1}E)^2 \tilde B, \ldots,({ A}^{-1}E)^{r}{\tilde B}\},  \quad \quad \quad \quad \quad \quad \quad \quad\quad \quad \quad \  (1)
\textrm{range}(W)=\textrm{span}\{C^T, E^T{ A}^{-T}C^T,(E^T{A}^{-T})^2C^T, \ldots
,(E^T{A}^{-T})^{r-1}C^T\}, \quad \quad (2)

where \tilde B=-A^{-1}B.

The transfer function \hat H of the reduced model has good approximation properties around s_0, which matches the first 2r moments of H(s) at s_0.

Using a set of k distinct expansion points \{s_1,\cdots,s_k\}, the reduced model obtained by, e.g.,


\textrm{range}(V)=\textrm{span}\{(A-s_1 {E})^{-1}E\tilde B,\ldots,(A-s_k {E})^{-1}E\tilde B   \},  \quad \quad \quad \quad \quad \quad \quad \quad (3)
\textrm{range}(W)=\textrm{span}\{E^T(A-s_1 {E})^{-T}C^T,\ldots,E^T(A-s_k {E})^{-T}C^T \},\quad \quad \quad \quad \quad (4)

matches the first two moments at each s_j, j=1,\ldots,k, see [3]. The reduced model is in the form as below

W^TE \frac{d{V z}}{t}=W^TA Vz +W^T B u(t), \quad \hat{y}(t)=CVz.

For the case of one expansion point in (1)(2), it can be seen that the columns of V, W span Krylov subspaces which can easily be computed by Arnoldi or Lanczos methods. The matrices V and W in (3)(4) can be computed with the rational Krylov algorithm in[3] or with the modified Gram-Schmidt process. In these algorithms only a few number of linear systems need to be solved, where matrix-vector multiplications are only used if using iterative solvers, which are simple to implement and the complexity of the resulting methods is roughly O(n r^2) for sparse matrices A, E.

References

  1. R.W. Freund, "Model reduction methods based on Krylov subspaces". Acta Numerica, 12:267-319, 2003.
  2. P. Feldmann and R.W. Freund, "Efficient linear circuit analysis by Pade approximation via the Lanczos process". IEEE Trans. Comput.-Aided Des. Integr. Circuits Syst., 14:639-649, 1995.
  3. 3.0 3.1 E.J. Grimme, "Krylov projection methods for model reduction. PhD thesis, Univ. Illinois, Urbana-Champaign, 1997.