W ESTFÄLISCHE

W ILHELMS -U NIVERSITÄT
M ÜNSTER

pyMOR
Model Order Reduction with Python

living knowledge
WWU Münster R. Milk, S. Rave PDESoft 2014
 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 2 /19

What is MOR
Discrete Problem
For given parameter µ ∈ P, find uµ,h ∈ Vh satisfying

K
X
θk (µ)Bk (uµ,h , vh ) = F (vh ) ∀vh ∈ Vh . (∗)
k=1

living knowledge
WWU Münster
René Milk ([email protected])
 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 2 /19

What is MOR
Discrete Problem
For given parameter µ ∈ P, find uµ,h ∈ Vh satisfying

K
X
θk (µ)Bk (uµ,h , vh ) = F (vh ) ∀vh ∈ Vh . (∗)
k=1

living knowledge
WWU Münster
I Assume that solving (∗) is very expensive.

René Milk ([email protected])

 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 2 /19

What is MOR
Discrete Problem
For given parameter µ ∈ P, find uµ,h ∈ Vh satisfying

K
X
θk (µ)Bk (uµ,h , vh ) = F (vh ) ∀vh ∈ Vh . (∗)
k=1

living knowledge
WWU Münster
I Assume that solving (∗) is very expensive.
I Need to solve for many µ ∈ P.

René Milk ([email protected])

 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 2 /19

What is MOR
Discrete Problem
For given parameter µ ∈ P, find uµ,h ∈ Vh satisfying

K
X
θk (µ)Bk (uµ,h , vh ) = F (vh ) ∀vh ∈ Vh . (∗)
k=1

living knowledge
WWU Münster
I Assume that solving (∗) is very expensive.
I Need to solve for many µ ∈ P.
I Offline-Phase: Use some fancy algorithm to
I compute snapshots S := {uµs ,h | s = 1, . . . , S}
I determine VN ⊆ span(S) with N = dim VN  dim Vh .

René Milk ([email protected])

 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 2 /19

What is MOR
Discrete Problem
For given parameter µ ∈ P, find uµ,h ∈ Vh satisfying

K
X
θk (µ)Bk (uµ,h , vh ) = F (vh ) ∀vh ∈ Vh . (∗)
k=1

living knowledge
WWU Münster
I Assume that solving (∗) is very expensive.
I Need to solve for many µ ∈ P.
I Offline-Phase: Use some fancy algorithm to
I compute snapshots S := {uµs ,h | s = 1, . . . , S}
I determine VN ⊆ span(S) with N = dim VN  dim Vh .
I Online-Phase: For new µ, solve (∗) restricted to VN .

René Milk ([email protected])

 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 2 /19

What is MOR
Reduced Problem
For given parameter µ ∈ P, find uµ,N ∈ VN satisfying

K
X
θk (µ)Bk (uµ,N , vN ) = F (vN ) ∀vN ∈ VN . (∗∗)
k=1

living knowledge
WWU Münster
I Assume that solving (∗) is very expensive.
I Need to solve for many µ ∈ P.
I Offline-Phase: Use some fancy algorithm to
I compute snapshots S := {uµs ,h | s = 1, . . . , S}
I determine VN ⊆ span(S) with N = dim VN  dim Vh .
I Online-Phase: For new µ, solve (∗) restricted to VN .

René Milk ([email protected])

 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 3 /19

What is pyMOR

I fast way of integrating MOR algorithms with existing

high-dimensional solvers

living knowledge
WWU Münster
René Milk ([email protected])
 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 3 /19

What is pyMOR

I fast way of integrating MOR algorithms with existing

high-dimensional solvers
I tool for interactive parameter studies

living knowledge
WWU Münster
René Milk ([email protected])
 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 3 /19

What is pyMOR

I fast way of integrating MOR algorithms with existing

high-dimensional solvers
I tool for interactive parameter studies

living knowledge
I Testbed for prototyping new methods

WWU Münster
René Milk ([email protected])
 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 3 /19

What is pyMOR

I fast way of integrating MOR algorithms with existing

high-dimensional solvers
I tool for interactive parameter studies

living knowledge
I Testbed for prototyping new methods

WWU Münster
I Free and Open Source Software (FreeBSD-2)

René Milk ([email protected])

 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 4 /19

pyMOR principles

I Generic algorithms on abstract interfaces

living knowledge
WWU Münster
,
,

René Milk ([email protected])

 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 4 /19

pyMOR principles

I Generic algorithms on abstract interfaces

I Modern OO design based on immutable types modeling
mathematical entities

living knowledge
WWU Münster
,
,

René Milk ([email protected])

 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 4 /19

pyMOR principles

I Generic algorithms on abstract interfaces

I Modern OO design based on immutable types modeling
mathematical entities

living knowledge
I Codebase tailored for interactive use and ease of

WWU Münster
experimentation

,
,

René Milk ([email protected])

 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 5 /19

pyMOR Building Blocks

I Vector arrays

living knowledge
WWU Münster
,
,

René Milk ([email protected])

 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 5 /19

pyMOR Building Blocks

I Vector arrays
I Operators

living knowledge
WWU Münster
,
,

René Milk ([email protected])

 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 5 /19

pyMOR Building Blocks

I Vector arrays
I Operators

living knowledge
I Discretizations

WWU Münster
,
,

René Milk ([email protected])

 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 6 /19

Batteries included

I Abstract grid interface with implementations for structured,

uniform triangle/quadrilateral cells designed for vectorized
operations with NumPy

living knowledge
WWU Münster
René Milk ([email protected])
 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 6 /19

Batteries included

I Abstract grid interface with implementations for structured,

uniform triangle/quadrilateral cells designed for vectorized
operations with NumPy
I stationary elliptic and instationary advection-diffusion

living knowledge
Discretizations and Operators

WWU Münster
René Milk ([email protected])
 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 6 /19

Batteries included

I Abstract grid interface with implementations for structured,

uniform triangle/quadrilateral cells designed for vectorized
operations with NumPy
I stationary elliptic and instationary advection-diffusion

living knowledge
Discretizations and Operators

WWU Münster
I glumpy and matplotlib based data visualization

René Milk ([email protected])

 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 7 /19

MOR(e) algorithms

I generic, VectorArrayInterface-based POD, Greedy and

Gram-Schmidt implementations
I automatic reduction of arbitrarily nested affine combinations

living knowledge
of operators

WWU Münster
,
,

René Milk ([email protected])

 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 8 /19

Parameter dependencies

I encapsulated in class Parameter, with a ParameterType

I model dependency by deriving from Parametric interface
I powerful automatic dependency injection method

living knowledge
I ParameterSpaceInterface as a model for P

WWU Münster
,
,

René Milk ([email protected])

 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 9 /19

Anatomy of a reduction process

living knowledge
WWU Münster
,
,

René Milk ([email protected])

 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 10 /19

Coupling external solvers I

I need to expose operators and vector data to pyMOR

living knowledge
WWU Münster
René Milk ([email protected])
 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 10 /19

Coupling external solvers I

I need to expose operators and vector data to pyMOR

I Possibilities include:
I Python bindings via pybindgen, Boost.Python, Python C-API, . . .
I Communication via sockets, MPI, . . .

living knowledge
I Data exchange through actual files

WWU Münster
René Milk ([email protected])
 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 10 /19

Coupling external solvers I

I need to expose operators and vector data to pyMOR

I Possibilities include:
I Python bindings via pybindgen, Boost.Python, Python C-API, . . .
I Communication via sockets, MPI, . . .

living knowledge
I Data exchange through actual files

WWU Münster
I Example: pymor/src/pymordemos/minimal_cpp_example

,
,

René Milk ([email protected])

 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 11 /19

Coupling external solvers II

living knowledge
WWU Münster
,
,

René Milk ([email protected])

 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 12 /19

Caching

I automatic and explicit method call caching using decorators

living knowledge
WWU Münster
,
,

René Milk ([email protected])

 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 12 /19

Caching

I automatic and explicit method call caching using decorators

I Immutable types allow automatic state-id generation

living knowledge
WWU Münster
,
,

René Milk ([email protected])

 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 12 /19

Caching

I automatic and explicit method call caching using decorators

I Immutable types allow automatic state-id generation
I swappable cache backends, in-memory and on-disk are

living knowledge
included

WWU Münster
,
,

René Milk ([email protected])

 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 13 /19

Live Demo

from multibat_experiments import *

p = params.best_shiquan_with_corner_1000s()
d = DuneDiscretization(**p)
d.parameter_space
u = d.solve((298,-0.0006))

living knowledge
d.visualize(u)

WWU Münster
,
,

René Milk ([email protected])

 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 13 /19

Live Demo

from multibat_experiments import *

p = params.best_shiquan_with_corner_1000s()
d = DuneDiscretization(**p)
d.parameter_space
u = d.solve((298,-0.0006))

living knowledge
d.visualize(u)

WWU Münster
pymor/src/pymordemos/thermalblock_gui.py –grid=140 2 2
3 3

,
,

René Milk ([email protected])

 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 14 /19

Tested

I Unittests with py.test

living knowledge
WWU Münster
René Milk ([email protected])
 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 14 /19

Tested

I Unittests with py.test

I Functional testing

living knowledge
WWU Münster
René Milk ([email protected])
 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 14 /19

Tested

I Unittests with py.test

I Functional testing

living knowledge
I leveraging Travis CI integration with GitHub

WWU Münster
René Milk ([email protected])
 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 15 /19

Users

I MULTIBAT: German Federal Ministry of Education and Research

project [1, 2]

living knowledge
WWU Münster
René Milk ([email protected])
 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 15 /19

Users

I MULTIBAT: German Federal Ministry of Education and Research

project [1, 2]
I dune-hdd via dune-pymor

living knowledge
WWU Münster
René Milk ([email protected])
 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 15 /19

Users

I MULTIBAT: German Federal Ministry of Education and Research

project [1, 2]
I dune-hdd via dune-pymor
I Practial course for lecture: Model Order Reduction for partial

living knowledge
differential equations

WWU Münster
René Milk ([email protected])
 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 15 /19

Users

I MULTIBAT: German Federal Ministry of Education and Research

project [1, 2]
I dune-hdd via dune-pymor
I Practial course for lecture: Model Order Reduction for partial

living knowledge
differential equations

WWU Münster
I Andreas Buhr: Interactive Simulations Using Parallel,
Event-Driven Localized Reduced Basis

René Milk ([email protected])

 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 16 /19

How to get pyMOR

I Ubuntu(12.04,13.10,14.04) packages:
$ apt-add-repository ppa:pymor/stable && apt-get
update && apt-get install python-pymor-demos
I via pypi into a virtualenv:
basically $ pip install pymor

living knowledge
but see http://pymor.org first

WWU Münster
I from source:
$ git clone https://github.com/pymor/pymor

,
,

René Milk ([email protected])

 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 17 /19

I Currently 3 active developers: Stephan Rave, Felix Schindler,

René Milk
I ~13000 lines of code (Ohloh)
I ~1600 commits
Contact us at http://listserv.uni-muenster.de/

living knowledge
I
mailman/listinfo/pymor-dev

WWU Münster
,
,

René Milk ([email protected])

 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 18 /19

Mario Ohlberger, Stephan Rave, Sebastian Schmidt, and Shiquan Zhang.

A model reduction framework for efficient simulation of li-ion batteries.
In Finite Volumes for Complex Applications VII-Elliptic, Parabolic and Hyperbolic Problems, pages 695–702.
Springer, 2014.

Stephan Rave.
BMBF-PROJEKT: MULTIBAT.
http://wwwmath.uni-muenster.de/num/ohlberger/research/projects/MULTIBAT/home.html?lang=de,

living knowledge
2014.

WWU Münster
,
,

René Milk ([email protected])

 W ESTFÄLISCHE
W ILHELMS -U NIVERSITÄT
M ÜNSTER pyMOR 19 /19

Thank you for

your attention.

living knowledge
WWU Münster
,
,

René Milk ([email protected])