## Learn to Use ITensor

# IQTensor

An IQTensor is similar to an ITensor in that it has named indices and supports most of the same operations. But an IQTensor also has a block sparse structure related to conservation of Abelian "quantum numbers". IQTensors actually have the same interface as ITensors, but the indices of an IQTensor are of type IQIndex.

IQTensors obey the rule that the only blocks allowed to be non-zero are those with the same QN "divergence". A block of an IQTensor corresponds to a specific sector of each IQIndex. The divergence of a given block is the sum of the QNs of the corresponding sectors times the Arrow direction of each IQIndex. On an intuitive level, the divergence of an IQTensor says how much the total quantum number will be changed by contracting with that IQTensor. As an example involving spins, an "Sz" operator has zero divergence while an "S+" operator has divergence +2 since it increases the total Sz by +2 (in units of spin 1/2).

Contracting IQTensors can be much more efficient than contracting ITensors. This is because IQTensors have an explicit sparse structure where many of its blocks are constrained to be zero.

Because IQTensor and ITensor have an identical interface (both implemented using the same template class), to see a list of class methods available for IQTensors see the ITensor documentation.

Functions acting on ITensors which are not discussed below can be assumed to have the same behavior for IQTensors.

IQTensor is defined in "itensor/iqtensor.h"; also see "itensor/iqtensor_impl.h". The IQTensor interface is defined in "itensor/itensor_interface.h".

## Synopsis

```
auto L = IQIndex("L",Index("L-",2),QN(-1),
Index("L0",4),QN( 0),
Index("L+",2),QN(+1));
auto S = IQIndex("S",Index("S-",1),QN(-1),
Index("S+",1),QN(+1));
//Create a zero IQTensor
auto A = IQTensor(L,S);
//Setting an element determines
//the divergence of this IQTensor
A.set(L(2),S(2),2.2);
//Compute IQTensor divergence
Print(div(A)); //prints: QN(0)
//Create an IQTensor with div QN(0)
//but otherwise random elements
auto B = randomTensor(QN(0),dag(L),S);
//Contracting IQTensors is very similar to
//contracting ITensors except contracted
//IQIndex's must have opposite Arrow directions
auto R = A * B; //contract over common IQIndex L
```

## IQTensor Class Methods Specializations

`.dag()`

Reverse the Arrow direction of all IQIndex's of this IQTensor and complex conjugate all the tensor elements.

## Operations on IQTensors

Most operation available for ITensors work similarly for IQTensors. The following are operations which have different behavior or are only defined for IQTensors:

`IQTensor * IQTensor -> IQTensor`

`IQTensor *= IQTensor`

Contract two IQTensors, summing over all matching IQIndex objects.

An IQIndex is only allowed to contract with a matching IQIndex of the opposite Arrow direction. If two IQIndex's match but have the same direction, the

`*`

operation throws an`ITError`

exception.`IQTensor + IQTensor -> IQTensor`

`IQTensor += IQTensor`

`IQTensor - IQTensor -> IQTensor`

`IQTensor -= IQTensor`

Add or subtract two IQTensors. Both IQTensors must have the same set of IQIndex's including Arrow directions. (However the order of the IQIndex's are not important.)

`IQTensor -> ITensor`

`toITensor(IQTensor T) -> ITensor`

Automatically convert an IQTensor to an ITensor: any IQTensor may be converted to an ITensor.

- Each Index of the resulting ITensor is the result of casting each IQTensor to its parent type, Index.
- The tensor elements remain unchanged except that the tensor becomes dense. Blocks of the IQTensor previously required to be zero are now allocated in memory and can be set to non-zero values.

The function

`toITensor`

is provided for users to better notate that a conversion to ITensor is the intended behavior.`IQTensor * ITensor -> ITensor`

`ITensor * IQTensor -> ITensor`

Contracting an IQTensor with an ITensor first converts a copy of the IQTensor to an ITensor as described above, then performs an ITensor contraction as usual.

`IQTensor += ITensor`

Add an ITensor whose set of Index objects correspond to a specific block of this IQTensor to that block. Each Index of the ITensor must be a block Index of an IQIndex of the IQTensor.

`IQTensor / IQTensor -> IQTensor`

`IQTensor /= IQTensor`

The non-contracting product is not currently implemented for IQTensors since it can give results with ill-defined divergence in certain cases.

## Functions for Constructing IQTensors

`randomTensor(QN q, IQIndex i1, IQIndex i2, ...) -> IQTensor`

`randomTensorC(QN q, IQIndex i1, IQIndex i2, ...) -> IQTensor`

Create an IQTensor having IQIndex's i1, i2, etc. and having divergence

`q`

. All non-zero blocks of this IQTensor are initialized with random elements.`randomTensorC`

is the same except blocks are initialized with random complex elements.`randomTensor(IQIndexVal iv1, IQIndexVal iv2, ...) -> IQTensor`

Given a set of IQIndexVals, return an IQTensor having IQIndex's corresponding to the IQIndexVals. The divergence of the IQTensor is determined by which block contains the element corresponding to the IQIndexVals. All blocks having this divergence are filled with random elements.

## Functions for Constructing Sparse IQTensors

`setElt(IQIndexVal iv1, IQIndexVal iv2, ...) -> IQTensor`

Return an IQTensor whose only non-zero element is the one corresponding to the provided set of IQIndexVals. This element has the value 1.0.

`delta(IQIndex i1, IQIndex i2, ...) -> IQTensor`

Return a diagonal-sparse IQTensor with the provided IQIndex's and all diagonal elements equal to 1.0. Because all diagonal entries are the same, uses only a constant amount of memory regardless of IQIndex size.

Contracting an IQTensor with a delta IQTensor is implemented through specialized routines for maximum efficiency and can be used, for example, to trace a pair of indices or replace one IQIndex with another.

`combiner(IQIndex i1, IQIndex i2, ...) -> IQTensor`

Return a sparse IQTensor with special "combiner" storage whose purpose is to combine the provided IQIndex's into a single larger IQIndex whose size is the product of

`i1.m()*i2.m()*...`

.The returned IQTensor has one extra IQIndex in addition to the IQIndex's provided. This extra IQIndex is the combined IQIndex.

Although combining IQIndex's in a naive way could result in many IQIndex blocks with the same quantum number, the IQTensor combiner takes extra steps to ensure that all degenerate QN sectors are "condensed" into just one sector for each possible QN.

To "uncombine" a combined IQIndex, just contract with the

`dag`

(Hermitian conjugate) of the original combiner.Click to Show Exampleauto A = IQTensor(I,J,K); //...set elements of A... auto C = combiner(K,I); //C has extra "combined" IQIndex Print(rank(C)); //prints: 3 //combine K and I into one IQIndex auto cA = A * C; Print(rank(cA)); //prints: 2 //"uncombine" IQIndex back into K and I auto uA = cA * dag(C); //resulting IQTensor is identical to A Print(norm(uA-A)); //prints: 0.0

## Functions for Analyzing IQTensors

`div(IQTensor T) -> QN`

Return the total divergence of this IQTensor.

Divergence is defined as follows:

- All non-zero blocks of an IQTensor have the same divergence.
- For any non-zero block, identify the sector of each IQIndex it corresponds to.
Compute the product of the QN of each IQIndex sector the block corresponds to times the Arrow direction of that IQIndex.

(Recall that Arrows can be either

`In`

or`Out`

; multiplying a QN by`In`

flips its sign while`Out`

leaves a QN's sign unchanged.)The divergence is the sum of these QN * Arrow products, which is again a QN.

`dir(IQTensor T, IQIndex I) -> Arrow`

Search each IQIndex of T for one that matches the IQIndex I (recall that Arrow directions are not used in IQIndex comparison). If a matching IQIndex is found, returns the Arrow direction of that IQIndex (as it appears on T).

`findIQInd(IQTensor T, Index i) -> IQIndex`

Searches each IQIndex of T, checking if the Index i is one of the "block indices" of that IQIndex. Returns that IQIndex if found.

If no IQIndex is found for which Index i is a block index, returns a default initialized IQIndex (recall that a default initialized IQIndex evaluates to

`false`

in a boolean context, so can be used to check whether a matching IQIndex was found).Click to Show Exampleauto lm = Index("L-",2); auto l0 = Index("L0",4); auto lp = Index("L+",2); auto L = IQIndex("L",lm,QN(-1), l0,QN( 0), lp,QN(+1)); auto S = IQIndex("S",Index("S-",1),QN(-1), Index("S+",1),QN(+1)); auto A = IQTensor(S,L); auto I = findIQInd(A,l0); if(!I) println("IQIndex not found"); Print(I == L); //prints: true

`isEmpty(IQTensor T) -> bool`

Returns true if the IQTensor has no blocks in its storage. Otherwise returns false.

`qn(IQTensor T, Index i) -> QN`

Searches each IQIndex of T until one is found which has the Index

`i`

as a block index. Returns the corresponding block QN of the IQIndex block labeled by`i`

.If no such IQIndex is found, throws an ITError exception.

`dir(IQTensor T, Index i) -> Arrow`

Searches each IQIndex of T until one is found which has the Index

`i`

as a block index. Returns the Arrow direction of that IQIndex.If no such IQIndex is found, throws an ITError exception.

## Developer / Advanced Methods

`mixedIQTensor(IQTensor i1, IQTensor i2,...) -> IQTensor`

Construct an IQTensor with IQIndex's i1, i2, etc. The IQTensor has QMixed storage, which allows any tensor component to be non-zero just like a regular ITensor. This means it has "mixed" QN divergence sectors.

The main purpose of mixedIQTensors is to create an ITensor but from a function that can only return one type; by choosing that return type to be

`IQTensor`

it is possible for the function to create both IQTensors and ITensors (which might not be well-defined IQTensors in general).

*This page current as of version 2.0.6*

Back to Classes

Back to Main