Styxis
Company
Knowledge

Hand off a sparse matrix exponential with its unitary residual

Derive the property and norm tolerance from the Hermitian generator. A returned matrix alone cannot admit a conserved-quantity calculation.

A property expected of the returned matrix

Compare the returned exponential’s unitary residual with the generator’s mathematical conditions and a prior tolerance before admitting a conserved-quantity calculation. Do not impose unitarity on generators that do not require it. Agree with downstream teams on the property the calculation assumes. Supply Truthound with externally calculated unitary residuals and approved norm budgets. [1, 2, 3]

The follow-up narrows the solve dispatch

SciPy #24366 reports a sparse expm result failing a unitary check at a particular scale while a dense control passes. Maintainer follow-ups confirm that the solve structure-detection fix in PR #24367 resolves the reported examples. That branch confirmation does not independently verify a particular installed release. [1]

Share the conjugate-transpose definition

Retain the Hermitian generator, scale, and calculation path. Externally compute the norm of U×U*−I, holding failures while comparing dense and sparse alternatives. First confirm that the generator requires unitarity and keep conjugate-transpose, norm, and scale definitions consistent. [1, 3]

A residual reference from the identity

The illustrative reference is the defining identity U×U*=I for a unitary matrix. The prior norm budget is 0.00000001. An observed residual 0 passes the <= comparison; 0.01 fails. This tolerance is a declared example, not a universal rule for every matrix size and norm. [1, 3]

Comparison after external matrix products

ColumnPairValidator in Truthound Python SDK 3.3.2 compares the residual externally calculated from U×U*−I with approved_residual_budget using <=. Operators supply complex matrix products, norms, and generator validation. Passing does not establish the full exponential or every physical conservation law; Core does not calculate expm. [1, 2, 3]

The resource cost of a dense alternative

Compare verified dense or alternative exponential paths with their memory costs, rechecking the same generator’s residual. Existing independent residual coverage reduces the benefit of another audit. Numerical agreement between two paths alone does not establish the required property. [1, 3]

Application conditions

Externally define the generator’s unitarity requirement, scale, conjugate transpose, norm, and prior budget.

Alternatives

Compare memory costs and same-generator residuals for verified dense or alternative exponential paths.

Limitations

  • Passing the residual does not establish the entire exponential or every physical conservation law.

Evidence

Event checked
Subject checked
Modified