11.07. Time discretization; the Euler family - I

Loading...

Del curso dictado por University of Michigan

The Finite Element Method for Problems in Physics

192 calificaciones

University of Michigan

The Finite Element Method for Problems in Physics

192 calificaciones

This course is an introduction to the finite element method as applicable to a range of problems in physics and engineering sciences. The treatment is mathematical, but only for the purpose of clarifying the formulation. The emphasis is on coding up the formulations in a modern, open-source environment that can be expanded to other applications, subsequently. The course includes about 45 hours of lectures covering the material I normally teach in an introductory graduate class at University of Michigan. The treatment is mathematical, which is natural for a topic whose roots lie deep in functional analysis and variational calculus. It is not formal, however, because the main goal of these lectures is to turn the viewer into a competent developer of finite element code. We do spend time in rudimentary functional analysis, and variational calculus, but this is only to highlight the mathematical basis for the methods, which in turn explains why they work so well. Much of the success of the Finite Element Method as a computational framework lies in the rigor of its mathematical foundation, and this needs to be appreciated, even if only in the elementary manner presented here. A background in PDEs and, more importantly, linear algebra, is assumed, although the viewer will find that we develop all the relevant ideas that are needed. The development itself focuses on the classical forms of partial differential equations (PDEs): elliptic, parabolic and hyperbolic. At each stage, however, we make numerous connections to the physical phenomena represented by the PDEs. For clarity we begin with elliptic PDEs in one dimension (linearized elasticity, steady state heat conduction and mass diffusion). We then move on to three dimensional elliptic PDEs in scalar unknowns (heat conduction and mass diffusion), before ending the treatment of elliptic PDEs with three dimensional problems in vector unknowns (linearized elasticity). Parabolic PDEs in three dimensions come next (unsteady heat conduction and mass diffusion), and the lectures end with hyperbolic PDEs in three dimensions (linear elastodynamics). Interspersed among the lectures are responses to questions that arose from a small group of graduate students and post-doctoral scholars who followed the lectures live. At suitable points in the lectures, we interrupt the mathematical development to lay out the code framework, which is entirely open source, and C++ based. Books: There are many books on finite element methods. This class does not have a required textbook. However, we do recommend the following books for more detailed and broader treatments than can be provided in any form of class: The Finite Element Method: Linear Static and Dynamic Finite Element Analysis, T.J.R. Hughes, Dover Publications, 2000. The Finite Element Method: Its Basis and Fundamentals, O.C. Zienkiewicz, R.L. Taylor and J.Z. Zhu, Butterworth-Heinemann, 2005. A First Course in Finite Elements, J. Fish and T. Belytschko, Wiley, 2007. Resources: You can download the deal.ii library at dealii.org. The lectures include coding tutorials where we list other resources that you can use if you are unable to install deal.ii on your own computer. You will need cmake to run deal.ii. It is available at cmake.org.

De la lección

11

In this unit, we study the unsteady heat conduction, or mass diffusion, problem, as well as its finite element formulation.

11.01. The strong form 16:29

11.01c In-Video Correction 0:43

11.02. The weak form, and finite-dimensional weak form - I 18:44

11.03. The weak form, and finite-dimensional weak form - II 10:14

11.04. Basis functions, and the matrix-vector weak form - I 19:54

11.04c In-Video Correction 0:44

11.05. Basis functions, and the matrix-vector weak form - II 12:03

11.05. Response to a question 0:51

11.06. Dirichlet boundary conditions; the final matrix-vector equations 16:57

11.07. Time discretization; the Euler family - I 22:37

11.08. Time discretization; the Euler family - II 9:55

11.09. The v-form and d-form 20:54

11.09ct.1. Coding assignment 4 - I 11:10

11.09ct.2. Coding assignment 4 - II 13:53

11.10. Analysis of the integration algorithms for first order, parabolic equations; modal decomposition - I 17:24

11.11. Analysis of the integration algorithms for first order, parabolic equations; modal decomposition - II 14:07

11.11c. In-Video Correction 1:06

11.12. Modal decomposition and modal equations - I 16:00

11.13. Modal decomposition and modal equations - II 16:00

11.14. Modal equations and stability of the time-exact single degree of freedom systems - I 10:49

11.15. Modal equations and stability of the time-exact single degree of freedom systems - II 17:38

11.15q. Response to a question 10:18

11.16. Stability of the time-discrete single degree of freedom systems 23:25

11.17. Behavior of higher-order modes; consistency - I 18:57

11.18. Behavior of higher-order modes; consistency - II 19:51

11.19. Convergence - I 20:49

11.20. Convergence - II 16:38

Conoce a los instructores

Krishna Garikipati, Ph.D.
Professor of Mechanical Engineering, College of Engineering - Professor of Mathematics, College of Literature, Science and the Arts

Welcome back.

We'll continue with our development of the finite

element method for these linear parabolic problems

in three dimensions where we have a single scalar unknown.

What we've managed to do in the last couple of segments is starting from

the weak form we've written out the stro, sorry, starting from the strong form we've

written out the weak form and arrived as far as the matrix vector equations, okay?

And if you have your notes in front of you, you can go back and

observe that the form of these final matrix vector equations was the following.

Right, the matrix vector equations.

Right, it takes on the following form.

We have this new matrix,

the mass matrix Md dot

plus Kd equals F.

Now in this form of the equations, we know that because of the,

there is no methods being employed.

The methods are on the weak form.

The the boundary conditions are already embedded in this set of matrix

equations, right?

What we do need to count for, however, is the, the initial condition.

Okay, so whereas we have this linear system of equations,

right, essentially an a, a first order of ODE, right?

So this is a first order ODE.

ODE being ordinary differential equation, so let me write that out.

Right, in our vector unknown, d, okay?

Now, because we've dealt with all this business about,

you know, number of nodes in the problem and

the number of Dirichlet boundary conditions and so on,

I'm going to, as we proceed from here on just say that d is

a vector living in an ns ndf dimensional space, right?

NDF is simply number of degrees of freedom, okay?

So, those are the number of degrees of freedom we are served.

NDF is the number of degrees of freedom we are actually solving for.

Now, this is the first order ODE, and what that implies is that we do need an initial

condition which is something we had stated in the strong form of the problem.

We just need to incorporate it here, and that ODE that we're working with i,

is the following, sorry, that dif, i,

initial condition that we need to work with is the following.

d at time to equals 0 equals

the vector of conditions

which are obtained by taking u

not at the particular location

of the corresponding node, okay?

And this is all the way down to u naught at x, now written in ndf, okay?

Right, instead of continuing to number by node, we've forgotten all about nodes.

Now we're just looking at this linear system of equations where d as an R

belongs to to an ndf lucrative space and this is what we have for

the initial condition, right?

It's convenient sometimes to also offer, to,

to refer to this vector as just d naught, okay?

Right, so this is the setting we have.

One thing I should mention is that whereas we sort

of carried through our development of the method

by considering a consistent mass matrix,

one can also do global lumping, okay?

So a globally lumped matrix, lumped mass matrix.

All right, let's denote it as M sub l for lumped, okay?

It can be defined as this thing can be defined.

Okay?

And the way to do that

is to simply say that

MAB lumped is equal to

the sum over c of MAC, okay?.

If A equals B, right, and

it is equal to 0, otherwise.

All right?

Okay, and

in some cases the problem is often solved with the globally lumped mass matrix.

Okay, so this is this is what we needed to complete from the previous from,

from the work, we set, set out in the previous segment.

We move on from here and begin looking at

the integration algorithms for this class of problems, okay?

And this is going to depend upon time discretization, okay?

The reason for it is that the the form of the matrix vector equation we're working

with is what we call a semi-discrete formulation.

And I explained why it is semi-discrete because we've

gone through the discretization in space, but don't get in time.

Okay, so now we do time discretization.

Now, in a move that may seem a little disappointing to some of you the time

discretization for this sort of problem is done with finite difference methods, okay?

However, the, the nature, the structure of the matrices M, K, and and

the vector F depend upon the finite element method.

So even though we're going to use finite difference methods in this

series of lectures for integration of time dependent equations,

time integration of time dependent equations.

Those methods are, do continue to be affected by the, by the underlying

finite element formulation for the spacial part of b d e, okay.

So this part is going to be finite difference.

Finite Difference methods, 'kay?

With that, that's what we're going to use.

Now, I'll make a remark here, which is that a fully, space-time,

Galerkin finite element method does exist.

Okay.

So, space-time.

Finite element methods do exist.

And they have remarkably good properties as well.

These types of methods are based on, carrying out integration.

Over, omega and

over the interval 0, T.

Okay?

So these are based actually on a formulation which is now going to look,

well, I, I, I, I probably shouldn't since, begin writing this formulation because

one, we'll get sucked into the rest of the development, but anyway,

they're based upon integrating over omega and over, the time integral.

And, they, they tend to have very good properties.

For instance, they have properties of the type where, the accuracy.

The accuracy with respect to time.

Is of higher order.

Then what we will be developing here.

Right?

Accuracy is of higher order than, with finite difference methods,

F D for finite difference.

They do come with a, they do come at a cost, though.

In particular, one has to develop a finite element mesh over space and time.

And this can pose a challenge when one is doing

problems that are 3D in space, right.

Because then you have a forth dimension in time,

which can get to be a bit of a challenge to visualize, definitely.

At any rate the methods exist and they're actually very sophisticated, okay.

So we get back now to our finite difference methods, right.

So the nature of the time discretization that we are,

we are seeking carry out is the following.

We are going to divide our time interval into sub-intervals.

So divide our time integral 0,T into sub-integrals.

Okay.

And these sub-intervals are,

0, let me do it this way.

These sub-integrals

are t0 to t1, t1 to t2.

So on until we come finally to.

TN minus 1,

to t N, okay.

So we have all of these sub-intervals, and

what we also imply here, of course.

Is that t0, or t nought is equal to 0,

and tN equals capital T.

Okay.

So, if you look at this what we've implied is that we've divided our total

time interval into capital N sub intervals, right.

So clearly here we have, N sub-intervals.

Okay?

The basis of our, of the methods that we will develop is the following, okay.

So what we are going to do here is consider.

A, consider an interval.

T sub little n,

to t sub n plus 1.

Okay?

All right.

Where, where, of course,

little n belongs to the integer integral 1,

2, N minus 1.

Sorry, 0 to N minus 1.

Okay.

So we consider an integral of this type.

The whole,

basis of our methods is going to be what is sometimes called time stepping, okay.

So, time stepping implies.

Right.

Time stepping implies that we know the solution at tn,

and we want to get to tn plus 1, okay.

Time stepping is essentially, is the following.

Knowing, the

solution at t sub n,

find it at tn plus 1.

Okay.

All right, so we're really, sort of, hopping along in time.

Okay, so if you think about it as, if you think about it as follows.

That's the time axis, right.

We've started here at 0, we want to get as far as t, right.

We have t0, t1, right.

We have tn, and we will have tn,

t little n plus 1, and here, of course,

we have t capital N, okay.

So the whole idea is that well, I know what, what, I know everything at tn.

How do I get to tn plus 1?

Okay, in order to proceed, and also actually in order to Later

on carry out some analysis, we need to lay down some more notation.

Okay?

And here is what we will do.

So some more notation.

Okay?

When I write d, at some time t n.

What I implied by this notation is the.

Solution vector d, if one were able to do exact integration in time.

Okay.

So this is the time exact solution, or

the time continuous solutions.

All right.

At time at t equals t n.

I am going to stick with the term time exactly.

Time continuous gets, gets a little bit more, ambiguous.

The time exact solution.

Okay?

And this simply means if we were able to integrate our ODE exactly.

Right? This is our matrix of vector ODE.

Right? If we were this with, with bound,

with initial conditions.

If we add a method to exactly integrate this,

then the solution that we would get at a given time t n would be this, okay.

In contrast, I will use d sub n,

okay, to denote the,

the algorithmic, solution at t n.

Okay?

So this is the algorithmic.

Solution, okay?

By which mean the solution that's obtained by applying some sort of discretized

time integration algorithm based upon the discretization that we've just spoken of.

Right?

The algorithmic solution obtained.

By the method to discrete,

to integrate

the discretized, the,

the time discretized

ODE, okay.

All right, so we have this notation.

What we need to do is

actually first of all say what we mean by the time discretized ODE.

And here's what we mean, okay?

Now if I were to simply re write out our ODE in the time-exact form,

here's what I, I, I would write it as m just as I have it up there, but

let me, explicitly put in the time.

Okay. Supposing we were looking at it time t n.

Okay? This would be m d dot at time t n plus

k d at time t n equals f which potentially

could also depend upon time.

All right.

Now when go to a time discretized version of this ODE,

what we will write here is the following.

We will write this out in the canonical form m.

Now instead of d dot t n, we will adopt the notation that,

you know, that well if d is our solution,

we can think of d dot as sort of the velocity of the solution, right.

Just, just as a term, it's really just the rate of the solution, but, but we use the,

the sort of canonical or the generalized velocity for a rate here.

And so we would write this as v, okay?

Indicating a velocity.

Now, this is not a time-exact quantity.

So we are going to denote it as v sub n, okay?

Using this sort of idea.

Okay?

Plus k d n

equals f at n.

Okay.

This is what we mean by saying that well here,

v at n is an, a discretize approximation,

It's a time discretized approximation again.

Time discretized, discretized

approximation of d dot right.

D n is of course just as we've defined above.

Okay, so this is what we call what I will refer to

as the discretized version of the ODE, okay.

So this is the time discretized

Are the time discrete, ODE.

All right?

Okay.

Explora nuestro catálogo

Inscríbete de manera gratuita y obtén recomendaciones personalizadas, actualizaciones y ofertas.

Coursera brinda acceso universal a la mejor educación del mundo, al asociarse con las mejores universidades y organizaciones, para ofrecer cursos en línea.

© 2018 Coursera Inc. Todos los derechos reservados.

Descargar en la App Store

Consíguelo en Google Play