Covariance and contravariance on manifolds

Explained also in covariance and contravariance in linear algebra.
If we have two charts for the same region of a manifold M, denoted x and y, related through a transformation

ϕ:xy

the matrix

dϕ=ybxa

is the one that converts the coordinates of the tangent space vectors in the base provided by x to the coordinates of the base provided by y.
But dϕ and dϕ1 are not just changes of coordinates, they can also be interpreted as a transformation of vectors (this occurs from the level of linear algebra, matrices can be base changes or transformation of space, if we leave the second base fixed). However, then, the matrix that transforms the base provided by the chart x into the base provided by y is not dϕ, but

dϕ1=xayb

Moreover, if to transform the coordinates of the vectors from chart x into the coordinates in y we have used column matrices and the matrix multiplied the vector from the left, to convert vectors from the base of chart x into those of the base of chart y we use row matrices (although internally they are vectors) and multiplication from the right. This can be switched to the comfortable form (columns and multiplication from the left) by taking the transpose.
So, the matrix to transform one into the other is:

d(ϕ1)t=(xayb)t

That is, vectors themselves transform in one direction, but their coordinates transform in the opposite direction. Hence the name contravariance.
It's somewhat analogous to what happens when we change the time. If we advance our clock, our temporal coordinate is at +1 but we are actually moving backwards (-1) because at the same hour as yesterday it is now earlier (there's more light).

Summing up:

Transformation Kind Expression
Coordinates change Contravariant ϕ:xy,y=ϕ(x)
Vector conversion Covariant {y}={x}xy (chain rule)
Vector components Contravariant Vy=yxVx
Covector conversion Contravariant {dy}=yx{dx} (chain rule)
Covector components Covariant ηy=ηxxy

Moreover:
Let's restrict to 2D case. When we have a vector, say

2x+y

and a covector

dx+3dy

we can represent them like an arrow, the first one, and a gradient (at least locally), the second one
The application of one into the other is the number of lines of the gradient that cross the vector. This is because to count the crossing lines we can count first in the horizontal direction and then add those from the vertical direction. See visualization of k-forms.
The gradient produced by, for example, dx+3dy corresponds to a line trough the point (0,1/3) and (1,0), its parallel line through the (0,0) and others at the same distance. Why the 3 produces 1/3? It has to do with real examples:

  1. imagine vectors are stock in a shop and covectors are prices, a cost, a barrier, to every product in the shop.
  2. a covector is a frequency, a vector is like a wavelength, and their product is analogous to a velocity (no estoy seguro de esto, es copiado de aquí y creo que se le puede sacar más partido a la analogía, viendo lo que pasa en la exponencial soluciión de la ecuación de onda multidimensional).

Observe that the set of parallel lines is perpendicular to the vector (arrow) (1,3)=x+3y. The similitud with the original dx+3dy is not coincidence: it is the covariant version of the other. It is exactly the same because in this case we are assuming the trivial metric

(1001)

What if we were with the metric

(4001)$$?Thecovector$dx+3dy$wouldbecomingfromthevector$1/4x+3y$,thatisstillorthogonaltotheparallellinebutwiththenewconceptoforthogonality.Nowconsiderachangeofbasis,aclockwiserotationofangle$θ$,$Rθ$,forexample.Itisapassivechange,inthesensethatwetransformourbasisbutthemaincharacters,thearrowandthegradient,arethesame.Thenewbasisisobtainedbyapplyingthematrix$Rθ$throughtherightsideorwiththetransposeof$Rθ$intheusualside(i.e.,iscovariant).Buthowiexpressednowthevector$2x+y$?Youcancheckthatyouhavetomultiply$Rθ$tothecomponents!Thisiswhyvectorsarecalledcontravariantvectors.Andwhataboutthegradient?Now,youcancheckthatthelineappearsrotatedandcorrespondtoadifferentcovector.Thenewcomponentscanbeobtainedbymultiplyingby$Rθ$intherightside,justlikethevectorconversion!Sotheyarecalledcovariantvectors.AndevenmoreaboutcontraandcovarianceInashop,imaginewehaveaproduct,apples,whoseamountwemeasureinkg.Thepriceismeasuredineur/kg.Whensomebodycomestotheshopandbuys4kgofappleswithapriceof25eur/kg,hedoesthesimplecomputation:

4 \cdot 2'5=10

toobtainanscalar.Itcouldbedonewithseveralitemsandseveralprices,butwealwaysobtainascalar.But,whatifwechangetheunitsinwhichwemeasuretheapples.Forexample,imaginethatapplesaresoldinpacketsof23kg.Observe:Theunititselfhasbeenmultipliedby23(kgtopacket).Thequantityofapplesboughtisnolonger4,butlessthan4.Itisdividedby23,forwhichwesayiscontravariant.Thepriceisnolonger25,butitchangetoagreaterprice.Itismultipliedby23.Soitiscalledcovariant.Thefactisthatweobtainthesamescalar:thetotalmoney.Everythingworkswellifwereplaceapplewithalistofproductsandalistofprices:vectorsandcovectors.