Welcome back, we're going to take the first steps towards analyzing our, ODE as original, using the Euler family, okay? And in order to carry out this analysis, there are couple of things that we have to do. The first is the following. Let us go back to our discretized ODE and write it out in a slightly different form, okay? The forms are completely equivalent to what we have seen before, the v and the d form. But for the purpose of analysis, let's do the following, okay? So the title of this segment is analysis, okay? And in particular, we are going to look at what I will call modal decomposition, Right? In order to get there, let's do the following. Let's write out the equation to these formulas, right? We are going to write it out as M, okay? We are going to directly write a time discretized approximation of finite difference approximation, if you will, of our time exact time derivative, right? So we're going to write d dot here as approximated as d n + 1- d at n / by delta t, Okay? So that's our approximation of M d dot. Now, the way the Euler family comes in is when we say that everything else is evaluated at n + alpha, All right? This is how we want to view it. Now, one more thing. We are interested in understanding what the basic properties of our integration algorithms are, what the basic properties are. And as often happens in the study of ODEs, we will take advantage of what is sometimes called a homogeneous form of the ODE, which is obtained by setting the forcing equal to zero, okay? So what we are aiming to analyze here is we want to analyze the stability, Stability and, What people define as the consistency of, The time, Integration, Algorithms, Right? We will define what we mean by their stability and consistency. But in order to do this, and especially if one thinks of the notion of stability, it makes sense to actually set the forcing equal to zero, okay? Think about why this may be, why it makes sense to set the force equal to zero if we want to start out by looking at stability, Okay? Now, it's simply because, well, how a solution evolves in time can depend upon the forcing, right? You give it a forcing, which makes it keep growing in time, well, it will keep growing, right? So we want to get that out of the picture, right? And instead, we want to understand the rigorous way to pose it and so to just ask the question that if there were no forcing, how does my time discretization and my introduction of this Euler family of algorithms effect the evolution of the problem, okay? So what we will look at is we will consider, The homogeneous, ODE, Okay, now the time, exact version of the homogeneous ODE that we're working with is this. M d dot + K d = 0, okay? With the initial condition d(0) = d 0, right? And the time discretized version of it is what we've written up here M (d m + 1- d n) / delta t + K d n + alpha, okay? = 0 with the 0 being given, Okay? So the first of those equations that I just showed is the time exact homogeneous ODE. And the next one is the time discretized version, using the Euler family, okay? One thing you will recognize here is that in all of this, what we are seeing is that, d n + alpha equals alpha and d n + 1- alpha, sorry, alpha d n + 1 + 1- alpha d n, okay? It's what we have. All right, now, in order to proceed, we are going to do the following. We're first going to look at stability, but even before that, what we want to do is carry out what we will call a modal decomposition of the problem, Right? In order to carry out a modal decomposition, we are going to take a step which is to invoke a related eigenvalue problem, okay? We will invoke, The related Generalized eigenvalue problem, Okay? And the generalized eigenvalue problem we want to invoke is the following. I'm going to write it as M, For a vector here. Let me not use d, but let me use phi, okay? Right, so the problem we want to look at is M phi = lambda K phi, Okay? Sorry, let me turn that around. Let me write it as, sorry, let me write it as this. Let me write it as lambda M phi = K phi, okay? This is a generalized eigenvalue problem. And perhaps, in a very obvious step, let me just change the left-hand side and the right-hand side, so we have K phi = lambda M phi, Okay, all right? So when you look at it in this form, if you haven't seen a generalized eigenvalue problem in the context of linear algebra before. You've seen a standard eigenvalue problem, probably. Where a standard eigenvalue problem is obtained by just setting M equalto what you would call the identity matrix in the corresponding dimension, okay? So let me just state this. So a remark is a standard eigenvalue problem, Is a standard eigenvalue problem would be of to form K phi = lambda phi, okay? And all of this of course, phi belongs to the Euclidean space or real space with number of dimension ndf, Okay, so what I've written down here will be a standard eigenvalue problem. A generalized eigenvalue problem is where instead of having the corresponding identity matrix here, right? We have, sorry, that's the isotropic tensor. Instead of having the identity matrix here, right, we have some other matrix here, right? And in our case it's a matrix n, which we know to be symmetric and positive definite, all right? Okay, so this is the generalized eigenvalue problem we want to consider, right? So it is this one that we will be working with, okay? Now, what one can do from here is the following, okay? What one can say is that for that problem, let say phi sub M, right, where M = 1, 2 ndf, okay? Let these be the eigenvectors, Okay, and let lambda M be the corresponding eigenvalue, Okay? All right, so lambda M is a corresponding eigenvalue. Now, what one can do is one can show that it's possible to construct a so-called orthonormalization of the the eigenvectors, okay? So the eigenvectors and eigenvalues satisfy the following equation. The eigenvectors and eigenvalues of course satisfy K phi m = lambda m times M phi m, right? Little m going form 1 to a number of degrees of freedom, all right? Okay, so given this, it is possible to construct. It's possible to choose the phi m, such that they satisfy a certain orthonormalize, a certain property of orthonormality, okay? In particular, the property for orthonormality that they satisfy is the following, the phi m, right, the set of Eigenvectors m = 1 to ndf, right? This set of eigenvectors can be orthonormalized, Okay, it can be orthonormalized to a different set psi sub m, okay? M = again to 1 to ndf, okay? It can be orthonormalized to a set psi n, which are such that this orthonormality property is an orthonormality with respect to m, okay? What that means is the following. If we take psi m, And dot it with M, the mass matrix, psi k, all right? This product is delta m k, okay? Where delta is our Kronecker delta, all right, okay? So this is the Kronecker delta, Okay, and furthermore, the set psi m, Right, that set of eigenvectors, I mean, I see they're orthonormalized. What I can further say is that they are M orthonormal, Right, it's simply, it can be interpreted as being orthonormal with respect to M, okay?