Continuous least squares is the version of least squares for functions, not just data points. In Linear Algebra and Differential Equations, it finds the best approximating function by minimizing the integral of the squared error over an interval.
Continuous least squares is a method for finding the best approximating function when your target is a continuous function, not a table of discrete data. Instead of minimizing error at a few points, you minimize the total squared error over an interval, usually written as an integral of the residual squared.
In this course, the idea shows up when you want to approximate a function using a simpler set of basis functions, such as polynomials or other functions you already know. You choose a form like c1φ1(x) + c2φ2(x) + ... + cnφn(x), then solve for the coefficient vector that makes the difference from the target function as small as possible in the least squares sense.
The setup uses the same core logic as ordinary least squares with matrices. The residual is the difference between the target function and the approximation, and the goal is to make its norm as small as possible. The condition for the best fit often leads to normal equations, where the residual is orthogonal to the space spanned by the basis functions.
A nice way to think about it is projection. You are projecting a function onto a subspace of allowed approximations, and the answer is the closest vector in that subspace under the inner product defined by an integral. That is why Gram matrices show up, since they collect the inner products of the basis functions.
A small example is approximating f(x) by a line a + bx on an interval like [0,1]. You pick a and b so the integral of (f(x) - (a + bx))^2 is as small as possible. If you set up the normal equations correctly, the coefficients come from solving a linear system, not from guessing the graph by eye.
The most common mistake is confusing continuous least squares with point-by-point curve fitting. Here, the error is measured across the whole interval, so a function can miss a few points and still be the best fit overall if its total squared error is smallest.
Continuous least squares matters because it connects differential equations, function approximation, and linear algebra in one tool. When an exact solution is hard or impossible to write down, this method gives you a controlled approximation instead of a vague estimate.
It also explains why inner products matter outside of vectors with coordinates. In this setting, functions behave like vectors, and the integral of their product acts like a dot product. That lets you use familiar linear algebra ideas, such as orthogonality, projection, basis functions, and Gram matrices, to solve approximation problems.
In a differential equations course, you may use this idea when a solution is too complicated for a closed form or when you want a smoother approximate solution on an interval. In linear algebra, it reinforces the idea that solving a least squares problem is really about finding the best approximating vector in a subspace.
If you can set up the residual, write the inner products, and solve the resulting normal equations, you can turn a messy function-fitting problem into a linear system you know how to handle.
Keep studying Linear Algebra and Differential Equations Unit 6
Visual cheatsheet
view galleryleast squares method
Continuous least squares is the function-based version of the least squares method. The big change is how the error is measured: instead of summing squared differences at listed data points, you integrate the squared residual over an interval. The goal is still the same, though, finding the best fit by minimizing total error.
Residuals
The residual is the gap between the target function and your approximating function. In continuous least squares, you square that gap and integrate it, so the residual is measured across the whole interval. If you do not write the residual correctly, the rest of the setup, including the normal equations, will be off.
Gram Matrix
The Gram matrix stores inner products of the basis functions you are using in the approximation. In continuous least squares, those inner products are usually integrals, so the matrix tells you how the basis functions relate to each other over the interval. It is the matrix version of the geometry behind the projection.
basis functions
Basis functions are the building blocks of the approximation space. In continuous least squares, you choose them first, then solve for the coefficient vector that gives the closest fit to the target function. Different basis choices change how easy the problem is to solve and how good the approximation looks.
A problem set question usually asks you to set up the integral for the squared error, choose a basis, and solve for the coefficient vector using normal equations. You may also need to interpret the result as a projection onto a subspace of functions. On quizzes, the trap is forgetting that the error is continuous, so you do not just match a few sample points. Instead, you work with inner products, residuals, and often a Gram matrix to build the linear system. If a differential equations problem asks for an approximate solution on an interval, continuous least squares gives you the coefficients that make the approximation best in the least-squares sense.
The least squares method is the general idea of minimizing squared error, often for discrete data in matrix form. Continuous least squares uses the same principle, but the error is measured with an integral over an interval, so it fits functions rather than just data points.
Continuous least squares finds the best approximating function by minimizing the integral of the squared residual over an interval.
It treats functions like vectors, so inner products, orthogonality, and projection still drive the solution.
The answer usually comes from normal equations, which turn the approximation problem into a linear system for the coefficients.
Basis functions and the Gram matrix tell you what space you are projecting onto and how the basis interacts with the interval.
If you see a continuous error measure instead of a discrete one, you are in continuous least squares, not ordinary point-by-point fitting.
It is a way to choose the best approximating function by minimizing the total squared error across an interval. Instead of checking only a few data points, you use an integral of the squared residual, which makes the method fit smoothly into function approximation problems.
You pick a set of basis functions, write the approximation as a linear combination of them, and form the squared error integral. Then you differentiate with respect to the coefficients or use orthogonality to get the normal equations. Solving that linear system gives the best fit.
They use the same minimizing idea, but they are not identical. Ordinary least squares usually works with discrete points and matrices built from data, while continuous least squares works with functions and inner products defined by integrals.
The Gram matrix records the inner products of the basis functions, and those inner products are usually integrals in this setting. It tells you how the basis functions overlap, which is exactly what you need when you solve the normal equations for the best approximation.