Understanding the abstract notion of an vector space

In summary, the abstract notion of a vector space is introduced by abstracting the properties of Euclidean vectors in \mathbb{R}^{n} (or \mathbb{C}^{n}), such as vector addition and scalar multiplication, to create a coordinate-free definition. This allows for the existence of vectors that are not necessarily geometric entities or limited to a finite number of dimensions. The concept of a vector possessing magnitude and direction is not necessary for a vector space, but can be added through an inner-product structure.
  • #1
"Don't panic!"
601
8
Hi,
I'm trying to justify to myself the abstract notion of a vector space and I would really appreciate if people wouldn't mind taking a look at my description and letting me know if it's correct, and if not, what is the correct explanation? :

"Vectors are most often introduced as ordered arrays of numbers (ordered "n-tuples") in [itex]\mathbb{R}^{n}[/itex] (or [itex] \mathbb{C}^{n}[/itex]), specifying their components along each coordinate line in Euclidean space. However, this viewpoint is very restrictive, as it requires one to introduce a specific coordinate system in order to describe a vector; an often non-trivial process, as it is not necessarily obvious which is the best coordinate system to choose. Vectors are intrinsically geometric entities, possessing both magnitude \& direction; they are mathematical objects that exist independently of any given coordinate system. Thus, we require a coordinate-free definition of a vector, and the 'space' that it lives in. We do this by abstracting the definition of a Euclidean vector in [itex]\mathbb{R}^{n}[/itex] (or [itex] \mathbb{C}^{n}[/itex]), retaining only the properties that are characteristic to the vectors themselves, and not a particular coordinate system. The characteristic properties that we refer to are the binary operations of vector addition and scalar multiplication."

Apologies in advance for the lack of mathematical rigour, I'm a physics student, but I am very interested in the mathematical formalism.
 
Physics news on Phys.org
  • #2
That's a good first step, but later on your will find there are vectors and vector spaces that are not "intrinsically geometric entities," and not limited to only a finite number of dimensions. For example a set of mathematical functions (e.g. all polynomials) can be considered as vectors that form a vector space. As an extreme example, you can have vector spaces where the "vectors" are themselves vector spaces.

You probably won't meet infinite dimensional vector spaces in a first course on abstract algebra or linear algebra, but they turn up in the maths of things like Fourier analysis, and in quantum mechanics.
 
  • #3
Thanks for the speedy response!
Is it more correct then to think of a "vector" as a mathematical object that satisfies a certain set of 'intrinsic' properties (the so-called vector-space axioms)?
I suppose I've always struggled to fully understand the abstraction from Euclidean vectors, visualised as arrows with magnitude & direction, to the more general notion of a vector. Is it purely that this restricts us to describing a vector as an ordered n-tuple in [itex]\mathbb{R}^{n}[/itex], dependent on a specific coordinate system, and we require a coordinate-independent definition, as there are many mathematical objects that satisfy the properties of vectors (such as addition and multiplication), all of which exist independently of any coordinate basis, thus requiring an abstracted definition?
Would you be able to explain this to me in a bit more depth? Thanks.
 
Last edited:
  • #4
"Don't panic!" said:
Is it more correct then to think of a "vector" as a mathematical object that satisfies a certain set of 'intrinsic' properties (the so-called vector-space axioms)?

Exactly. If it walks like a duck and quacks like a duck, it's a duck. The same principle applies to vector spaces, and almost everything else in "advanced" math.

I suppose I've always struggled to fully understand the abstraction from Euclidean vectors, visualised as arrows with magnitude & direction, to the more general notion of a vector. Is it purely that this restricts us to describing a vector as an ordered n-tuple in [itex]R^{n}[/itex], dependent on a specific coordinate system, and we require a coordinate-independent definition.

Suppose you have some physical system that you want to model using vectors - e.g. the forces, accelerations, momentum, etc, in a dynamics problem. The point is that the physics is independent of the coordinate system you use to model it. To solve a particular problem, you can often pick a coordinate system that makes the algebra simple, but the answer would have the same physical meaning whatever coordinate system you used. The physical system doesn't "know" anything about your choice of coordinate system, it just does what it does.

The basic equations of physics (Newton's second law, Maxwell's equations, etc) often involve vectors, but they don't involve coordinate systems. The same is true of the basic theorems in vector calculus, like the divergence theorem. That theorem connects what happens inside a region in space, with what happens on the boundary of the region. It doesn't matter what shape the region is, nor does it matter how you choose to define the shape using coordinates or anything else.

The idea of representing a vector as "an arrow on a diagram" instead of "its x,y,z coordinates" goes some way towards the idea that the physics is independent of the coordinate system, but of course you can't do much with an arrow on a diagram, until you make it into a mathematical concept.

If you have some time for self-study, it might help to start where most pure math courses start on "abstract algebra", which is with groups rather than vector spaces. A group is a set of objects with only one mathematical operation (and only 4 axioms). An example is the set of integers (positive and negative) and addition. But there are many "real world" examples of different types of things which have the properties of a group, apart from "numbers" and "arithmetic".

Reading a bit of group theory might help to get used to the idea of why more general abstract definitions are useful. It's harder to find "simple" examples of vector spaces, where Cartesian coordinates are not the "obvious" way to represent them.
 
  • #5
Ah, ok. Thanks very much, your comments have been very helpful. I shall look into reading up on group theory.

One more quick question, possibly a little stupid (so apologies for that). Do vectors always possesses magnitude and direction in their abstracted definition?
 
Last edited by a moderator:
  • #6
"Don't panic!" said:
One more quick question, possibly a little stupid (so apologies for that). Do vectors always possesses magnitude and direction in their abstracted definition?

There's an extra structure called an inner-product that can be attached to a vector space to recover some of these properties, but it's not necessary for a vector space to have them.
 
  • #7
Ok, so in general vectors do not have magnitude and direction, but if they form an 'inner-product space', then these characteristics can be recovered? Is the notion of a direction of a given vector only introduced once one specifies a particular coordinate system?
 
  • #8
The best way I can explain vectors is linear combinations. If you can take linear combinations of something, it's a vector space. i.e. αx+βy makes sense. Commutivity, associativity and distributivity etc. are all axioms, meaning they must satisfied to be called a "vector space" BUT in general these things are trivial. The only time you'll really worry about proving something satisfies these axioms is in your linear algebra class.

Polynomials, continuous functions, matrices, Euclidean vectors, etc. are all vector spaces because you can take linear combinations of them.

In general though, this is just an example of an algebraic structure. A set with one or more operations on it. If you get to abstract algebra, you learn about all sorts of structures. Groups, rings, fields, algebras, modules, vector spaces, latices.

"Don't panic!" said:
Ok, so in general vectors do not have magnitude and direction, but if they form an 'inner-product space', then these characteristics can be recovered? Is the notion of a direction of a given vector only introduced once one specifies a particular coordinate system?

Inner product spaces define a notion of angle or orthogonality. Euclidean space with the dot product is an inner product space. if x*y=0, the vectors are orthogonal. This might not be a big deal to you now, but it will be as you move on.

This will probably be beyond you for right now, but in functional analysis (used in quantum a lot; practically invented for it), you discuss metric spaces, normed vector spaces, and inner product spaces. A metric space is ANY set (doesn't have to be a vector space) with a notion of "distance". A normed vector space is a vector space with a notion of "length". An inner product space is a vector space with a notion of "angle".

The cool thing is that if you have an IP space, you get a NV space, which gives you a metric space. The other way doesn't hold. If you have a metric space, you might not have a NV space (you might not even have a vector space). If you have a NV space, you might not have an inner product space.

With general metric spaces and NV spaces, you have a lot of weird examples. Distances and lengths that don't match up with reality. But in order to have an IP space, it really restricts what you can have and makes things match up reality much better.

This might be beyond you for now, but I guess what I'm trying to say is that inner products on a space REALLY wrangle it, make things more "normal".

Edit: for your question of "magnitude", a normed vector space gives a notion of magnitude (or length).
 
Last edited by a moderator:
  • #9
Thanks johnqwertyful! A very insightful description, this has given me a lot more understanding than I had when it was first introduced to me. Much appreciated!
Is there any intrinsic property though, that gives a sense of "direction" to a vector ?
 
Last edited:
  • #10
Sorry, I did not respond to that.

Really, what is direction? If you were to say that a force is acting in the "x-direction", what does that mean? Well it means that the "angle" between the force vector, F, and the x direction, i, is 0. If you have an inner product space, you have a notion of "angle". You can define an angle between two vectors as: http://upload.wikimedia.org/math/8/9/1/891f5665ee88157f3f54f28fe8633b24.pngTo answer your question, in the general case vector spaces don't have a notion of magnitude or length. But if you add a norm structure, you get a magnitude. If you add an inner product, you get a notion of direction. But an inner product implies a norm.

Also, an inner product really restricts what norms are available. Only certain norms can become inner products. So really everything boils down to an inner product. Inner product spaces are really well behaved.
 
  • #11
Great. Thanks, this has really helped.

In this abstracted framework, is it then correct to distinguish between vectors and scalars by their behavior under coordinate transformations, i.e. a scalar quantity is invariant under coordinate transformations, whereas a vector transforms in such a way as to preserve its overall form?
 
Last edited:
  • #12
Also, just as a side note, you can think of the angle definition as a result of the Cauchy-Schwarz inequality in inner product spaces.
 
  • #13
"Don't panic!" said:
Is there any intrinsic property though, that gives a sense of "direction" to a vector ?
In the most abstract? No. The concepts of magnitude and direction are not a part of the concept of an abstract vector space.

Adding the concept of a norm (and not all vector spaces have a norm) gives the concept of magnitude. Some function spaces don't even admit the concept of a norm, but they are nonetheless vector spaces. It's only when the norm is induced by an inner product that it makes sense to talk about "magnitude" and "direction". The Manhattan norm (e.g., "drive two blocks north then three blocks east") is an example of a normed vector space that doesn't admit the concept of direction.
 
  • #14
"Don't panic!" said:
Great. Thanks, this has really helped.

In this abstracted framework, is it then correct to distinguish between vectors and scalars by their behavior under coordinate transformations, i.e. a scalar quantity is invariant under coordinate transformations, whereas a vector transforms in such a way as to preserve its overall form?

In the language of vector spaces, "coordinates" give way for a basis. Formally a basis is a "linearly independent" set of vectors that "span" the space. In other words a basis is a set of vectors such that every vector in the space can be UNIQUELY written as a linear combination of the vectors.

A basis for R^3 is i, j, k. Meaning every vector in R^3 is uniquely ai+bj+ck. Coordinates may make sense in the abstract sense, but it's mainly a geometry thing.

There ARE change of bases. I'm not quite sure what you mean by "scalar quantity is invariant".
 
  • #15
"Don't panic!" said:
Great. Thanks, this has really helped.

In this abstracted framework, is it then correct to distinguish between vectors and scalars by their behavior under coordinate transformations, i.e. a scalar quantity is invariant under coordinate transformations, whereas a vector transforms in such a way as to preserve its overall form?
What do you mean by "a vector transforms in such a way as to preserve its overall form?"

Consider the set of all real numbers of the form ##p + \sqrt{2} q##, where ##p## and ##q## are rational numbers. It's straightforward to show that this is a 2-dimensional vector space over the rationals. The obvious basis for this vector space is ##\{1, \sqrt{2}\}##. Another basis is ##\{1 - \sqrt{2}, 1 + \sqrt{2}\}##. All that a coordinate transformation does is to express ##p + \sqrt{2} q## in the alternate form ##r (1 - \sqrt{2}) + s(1 + \sqrt{2})##, where ##r = (p-q)/2## and ##s = (p+q)/2##. But the vector (real number) hasn't changed in any way.

By the way, the above example also gives you a concrete case of a vector space where there's no useful notion of direction/angle. All the vectors lie on the same real line, so they are all parallel with each other.
 
  • #16
jbunniii said:
What do you mean by "a vector transforms in such a way as to preserve its overall form?"

Sorry, not well worded - I meant as in the vector itself exists independently of any coordinate system, so its "shape" should not change under coordinate transformations (its components should vary in such a way, that they compensate for any change in the basis that it is represented in), right?

Sorry, I haven't explained my problem very well re. scalars. I guess the crux of my issue is, is that I am struggling to place what a 'scalar' quantity is in the framework of abstract vector spaces?
 
  • #17
"Don't panic!" said:
I guess the crux of my issue is, is that I am struggling to place what a 'scalar' quantity is in the framework of abstract vector spaces?
It's an element of a "field". A field in mathematics is (informally) a mathematical structure that has addition, subtraction, multiplication, and division.
 
  • #18
can one view elements of a field as 0-dimensional 'numbers' (what really distinguishes them from vectors, as they seem to obey the vector space axioms as well). Sorry, all this must seem very trivial, I'm just struggling a bit to transition my understanding from the specific definition of a scalar in the physical sense (i.e. a quantity that is completely defined by its magnitude, such that it is invariant under rotations, translations, etc. in a given coordinate system).

...is it that in a given coordinate system, performing an 'active transformation' on a given vector, will in general, change the form of that vector in some way, whereas, doing the same procedure to a scalar will leave it unchanged?
 
Last edited by a moderator:
  • #19
"Don't panic!" said:
Sorry, not well worded - I meant as in the vector itself exists independently of any coordinate system
Correct.
so its "shape" should not change under coordinate transformations (its components should vary in such a way, that they compensate for any change in the basis that it is represented in), right?
I think you have the right idea, but your wording is a bit unconventional. The vector is some object which may not have a "shape" in the geometric sense of the word. But the object, whatever it is, does not change regardless of what basis you use to represent it.

Sorry, I haven't explained my problem very well re. scalars. I guess the crux of my issue is, is that I am struggling to place what a 'scalar' quantity is in the framework of abstract vector spaces?
You can add vectors together to get new vectors, but in general you cannot multiply vectors together. However, you can multiply them by scalars. Geometrically in ##\mathbb{R}^n##, this has the effect of changing the length of the vector but not its direction (except possibly a sign change). Thus if we fix a specific nonzero vector ##v##, then the set ##\{\alpha v : \alpha \in \mathbb{R}\}## consisting of all scalar multiples of ##v## is a line through the origin, parallel with ##v##.

This notion carries over to a general vector space ##V## over a field of scalars ##F##: given a nonzero vector ##v \in V##, the set ##\{\alpha v : \alpha \in F\}## is a one-dimensional subspace of ##V##, which we call the subspace spanned by ##v##. This subspace includes the origin (zero vector) but it doesn't necessarily look like a line. For example, in the vector space I indicated previously, ##\{p + \sqrt{2}q : p,q \in \mathbb{Q}\}##, the subspace generated by ##\sqrt{2}## is the set of all rational multiples of ##\sqrt{2}##. Another example: if the vector space is ##\mathbb{C}^2## and the scalars are ##\mathbb{C}##, then the subspace generated by ##(1,0)## looks like the entire complex plane.
 
Last edited:
  • #20
"Don't panic!" said:
can one view elements of a field as 0-dimensional 'numbers' (what really distinguishes them from vectors, as they seem to obey the vector space axioms as well). Sorry, all this must seem very trivial, I'm just struggling a bit to transition my understanding from the specific definition of a scalar in the physical sense (i.e. a quantity that is completely defined by its magnitude, such that it is invariant under rotations, translations, etc. in a given coordinate system).
The real numbers are a 1-dimensional vector space over themselves. Dimension has a very precise definition. The dimension of a vector space is how many vectors it takes to form a basis.
The only vector space with dimension 0 is simply the identity, {0}.

Also, like I said coordinates aren't really used. Bases are.
 
  • #21
I think you have the right idea, but your wording is a bit unconventional. The vector is some object which may not have a "shape" in the geometric sense of the word. But the object, whatever it is, does not change regardless of what basis you use to represent it.
...Yes, that is what I was alluding to, just couldn't think of how to word it at the time, thanks for neatening it up.

The real numbers are a 1-dimensional vector space over themselves. Dimension has a very precise definition. The dimension of a vector space is how many vectors it takes to form a basis.
The only vector space with dimension 0 is simply the identity, {0}.

Also, like I said coordinates aren't really used. Bases are.

Would it be be correct then to say, that a scalar is a mathematical object that is an element of a field that pertains to a particular vector space. In a sense, each element "scales" the vectors in the overlying vector space?
Also, would it be correct to say within a given basis, what distinguishes a scalar from a vector is that a scalar will remain unchanged if it is, for example, rotated or translated within that basis, whereas a vector, in general, will change in some way upon such a rotation or translation within a given basis?

Thanks to all of you for all your help.
 
  • #22
"Don't panic!" said:
Would it be be correct then to say, that a scalar is a mathematical object that is an element of a field that pertains to a particular vector space. In a sense, each element "scales" the vectors in the overlying vector space?
Yes, that's reasonable.
Also, would it be correct to say within a given basis, what distinguishes a scalar from a vector is that a scalar will remain unchanged if it is, for example, rotated or translated within that basis, whereas a vector, in general, will change in some way upon such a rotation or translation within a given basis?
No, that would not be correct. Scalars and vectors are two different kinds of objects. They are distinguished by the fact that vectors are elements of the vector space, whereas scalars are not: they are elements of the field associated with the vector space. It doesn't make sense to talk about rotating or translating the scalars. (Indeed, in general it doesn't make sense to talk about rotating or translating the vectors, either. For example, if you have a vector space whose elements are functions, what does it mean to rotate the function?)
 
  • #23
No, that would not be correct. Scalars and vectors are two different kinds of objects. They are distinguished by the fact that vectors are elements of the vector space, whereas scalars are not: they are elements of the field associated with the vector space. It doesn't make sense to talk about rotating or translating the scalars. (Indeed, in general it doesn't make sense to talk about rotating or translating the vectors, either. For example, if you have a vector space whose elements are functions, what does it mean to rotate the function?)

Yes, thanks for pointing this out, I realized soon after writing the post that this didn't really make sense in general; I think I've caused confusion for myself by trying to conflate the physical and mathematical definitions of a scalar!
 
  • #24
"Don't panic!" said:
Yes, thanks for pointing this out, I realized soon after writing the post that this didn't really make sense in general; I think I've caused confusion for myself by trying to conflate the physical and mathematical definitions of a scalar!

Scalars are almost always going to be the real or complex numbers. It's the same in math and physics. Very rarely do you consider exotic fields.

But the definitions are the same in math and physics. A scalar "scales" the vector. Whether the vector is an arrow or function or matrix or whatever.

In physics, believe it or not, you WILL use the notion of an abstract vector space, particularly in QM.
 
  • #25
johnqwertyful said:
Scalars are almost always going to be the real or complex numbers. It's the same in math and physics. Very rarely do you consider exotic fields.

This is a bit of an overstatement. In some areas of math you will deal almost exclusively with vector spaces over more exotic fields. I imagine the same is increasingly true in parts of theoretical physics, but being largely ignorant of such matters, I cannot say for sure.
 
  • #26
johnqwertyful said:
Scalars are almost always going to be the real or complex numbers. It's the same in math and physics. Very rarely do you consider exotic fields.

But the definitions are the same in math and physics. A scalar "scales" the vector. Whether the vector is an arrow or function or matrix or whatever.

In physics, believe it or not, you WILL use the notion of an abstract vector space, particularly in QM.

Sorry, yes I have actually used the notion of an abstract vector space on my QM course.

I just find it a bit confusing as I was always taught (before being introduced to the notion of abstract vector spaces) that scalars are quantities that are invariant under rotations and translations (in a given basis)?!

I get that scalars are the elements of a given underlying field describing a given vector space, but it is the above description that I am finding difficult to grasp in terms of abstract vector spaces (apologies if this sounds stupid).
 

What is a vector space?

A vector space is a mathematical structure that consists of a set of vectors and operations that can be performed on those vectors, such as addition and scalar multiplication. It is a fundamental concept in linear algebra and has applications in many areas of science and engineering.

What are the defining properties of a vector space?

There are 10 defining properties of a vector space, but the most important ones are closure under addition and scalar multiplication, and the existence of a zero vector and additive inverses. These properties ensure that the operations on vectors behave in a consistent and predictable manner.

What is the difference between a vector space and a subspace?

A subspace is a subset of a vector space that also satisfies the 10 defining properties. In other words, it is a smaller vector space contained within a larger vector space. All vector spaces have at least two subspaces - the trivial subspace consisting of only the zero vector, and the entire vector space itself.

How do vector spaces relate to linear transformations?

Vector spaces and linear transformations are closely related. A linear transformation is a function that maps one vector space to another, while preserving the vector space structure. In other words, the operations on vectors in the original space will also hold in the transformed space. This makes linear transformations a powerful tool for studying vector spaces.

What are some real-world applications of vector spaces?

Vector spaces have a wide range of applications in science and engineering. They are used in physics to describe quantities such as velocity and force, in computer graphics to represent shapes and movements, in economics to model supply and demand, and in machine learning to analyze and classify data. Essentially, any situation that involves multiple quantities that can be added and multiplied together can be modeled using vector spaces.

Similar threads

  • Linear and Abstract Algebra
Replies
3
Views
177
  • Linear and Abstract Algebra
Replies
6
Views
755
  • Linear and Abstract Algebra
Replies
2
Views
1K
  • Differential Geometry
Replies
21
Views
586
  • Linear and Abstract Algebra
Replies
3
Views
1K
  • Linear and Abstract Algebra
Replies
1
Views
813
  • Linear and Abstract Algebra
Replies
32
Views
2K
  • Linear and Abstract Algebra
Replies
3
Views
1K
  • Linear and Abstract Algebra
Replies
8
Views
4K
  • Linear and Abstract Algebra
Replies
14
Views
3K
Back
Top