nLab
trace

Traces

Definition

If a is a dualizable object in a symmetric monoidal category C, there is a notion of the trace of an endomorphism f:aa, which reproduces the ordinary notion of trace of a linear map of finite dimensional vector spaces for the case that C=Vect.

The idea of the trace operation is easily seen in string diagram notation: essentially one takes the endomorphism afa, “bends it around” using the duality and the symmetry and connects its output to its input.

1 tr(f) 1:= 1 a * a Id a * f a * a b a *,a a a * 1\array{ 1 \\ \;\;\;\downarrow^{tr(f)} \\ 1 } \;\;\; := \;\;\; \array{ & 1 \\ & \downarrow \\ a^* &\otimes& a \\ \downarrow^{\mathrlap{Id_{a^*}}} && \;\;\downarrow^f \\ a^* &\otimes& a \\ & \downarrow^{\mathrlap{b_{a^*, a}}} \\ a &\otimes& a^* \\ & \downarrow \\ & 1 }

This definition makes sense in any braided monoidal category, but often in non-symmetric cases one wants instead a slightly modified version which requires the extra structure of a balancing.

The trace of the identity 1 a:aa is called the dimension or Euler characteristic of a.

Examples

  • C=Vect with its standard monoidal structure (tensor product of vector spaces): in this case tr(f) is the usual trace of a linear map;

  • C=SuperVect=(Vect 2,,b), the category of 2-graded vector spaces with the nontrivial symmetric braiding which is 1 on two odd graded vector spaces: in this case the above is the supertrace on supervectorspaces, str(V)=tr(V even)tr(V odd).

  • C=Span(Top op): here the trace is the co-span co-trace which can be seen as describing the gluing of in/out boundaries of cobordisms

  • C=Span(Grpd): this reproduces the notion of trace of a linear map within the interpretation of spans of groupoids as linear maps in the context of groupoidification and geometric function theory, made explicit at span trace

Categorification

See trace of a category

Partial trace

If the morphism described above is the endomorphism of a tensor product object VW, then there is a similarly evident way to “bend around” only the W-strand.

TO DO: Draw the diagram just described.

Matrix representation

Suppose V, W are finite-dimensional vector spaces over a field, with dimensions m and n, respectively. For any space A let L(A) denote the space of linear operators on A. The partial trace over W, TrW, is a mapping

TL(VW)Tr W(T)L(V).T \in L(V \otimes W) \mapsto Tr_{W}(T) \in L(V).
Definition

Let e 1,,e m and f 1,,f n be bases for V and W respectively. Then T has a matrix representation {a kl,ij} where 1k,im and 1l,jn relative to the basis of the space VW given by e kf l. Consider the sum

b k,i= j=1 na kj,ijb_{k,i} = \sum_{j=1}^{n}a_{k j,i j}

for k,i over 1,,m. This gives the matrix b k,i. The associated linear operator on V is independent of the choice of bases and is defined as the partial trace.

Example

Consider a quantum system, ρ, in the presence of an environment, ρ env. Consider what is known in quantum information theory as the CNOT gate:

U=0000+0101+1110+1011.U={|00\rangle}{\langle 00|} + {|01\rangle}{\langle 01|} + {|11\rangle}{\langle 10|} + {|10\rangle}{\langle 11|}.

Suppose our system has the simple state 11 and the environment has the simple state 00. Then ρρ env=1010. In the quantum operation formalism we have

T(ρ)=12Tr envU(ρρ env)U =12Tr env(1010+1111)=1100+11112=11T(\rho) = \frac{1}{2}Tr_{env}U(\rho \otimes \rho_{env})U^{\dagger} = \frac{1}{2}Tr_{env}({|10\rangle}{\langle 10|} + {|11\rangle}{\langle 11|}) = \frac{{|1\rangle}{\langle 1|}{\langle 0|0\rangle} + {|1\rangle}{\langle 1|}{\langle 1|1\rangle}}{2} = {|1\rangle}{\langle 1|}

where we inserted the normalization factor 12.

References

For partial trace, particularly its application to quantum mechanics, see:

  • Nielsen and Chuang, Quantum Computation and Quantum Information