QTechTheory/QuESTlink: v0.14

This minor release adds support for diagonal matrices, a validation improvement, and a patch to an esoteric bug in ApplyCircuitDerivs[].

• Gates U, UNonNorm and Matr, which previously accepted dense matrices (specified element-wise), now addtionally accept diagonal matrices specified as flat lists. Such diagonal gates have bespoke backend QuEST simulation functions (invoked by functions like ApplyCircuit[]), and bespoke/optimised Mathematica handling (invoked by functions like CalcCircuitMatrix[] and SimplifyCircuit[]).

  For example, operators specified as

  U_0[ {{a,0}, {0, b}} ]
  

  and

  U_0[ {a,b} ]
  

  are equivalent, though the latter is (approximately) quadratically faster (in terms of its dimension) to simulate.

• derivative functions ApplyCircuitDerivs[], CalcMetricTensor[] and CalcExpecPauliStringDerivs[] now explicitly check whether the given density matrix is Hermitian (otherwise throwing an error), since this is a pre-condition to their correct function.

• patched an esoteric ApplyCircuitDerivs bug differentiating w.r.t a parameter in a Matr[] operator upon a density matrix. Precisely; the existing Matr gate modifies density matrices rho under rho -> Matr rho

  Previously, ApplyCircuitDerivs[] would compute the derivative of Matr as D( Matr ||rho>> ) = conj(Matr) (x) D(Matr) ||rho>> + adj where ||rho>> notates the equivalent vector under the Choi isomorphism.

  This formulation is erroneous, and treats Matr like a unitary gate operated as rho -> U rho dagger(U)

  Instead, the true derivative of Matr is simply; D(Matr) ||rho>>

  This bug was extremely esoteric and was likely not encountered by users, since such state derivatives are typically used by variational algorithms leveraging unitary ansatzse. It also did not affect similar functions like CalcExpecPauliStringDerivs[] and CalcMetricTensor[], which already forbid non-trace-preserving operators like Matr.