The Polar Decomposition of a Kraus Operator: Effect, Backaction and the Measurement Boundary

Introduction

A Kraus operator is the amplitude of one outcome, not the outcome itself. The probability of the outcome is the trace of the effect $\tilde M_{\mathrm{kraus}}^{*}\tilde M_{\mathrm{kraus}}$; the post-measurement state is the normalized image $\tilde M_{\mathrm{kraus}}\tilde{Ϛ}_{\Psi}\tilde M_{\mathrm{kraus}}^{*}$; and the element $\tilde M_{\mathrm{kraus}}$ carries both, together with a phase that the probability discards. This article asks what the polar decomposition of $\tilde M_{\mathrm{kraus}}$ says about the three objects.

The thesis is one sentence. Every Kraus operator factors as $\tilde M_{\mathrm{kraus}}=\tilde M_{\mathrm{kraus},H}\tilde M_{\mathrm{kraus},0}$, with $\tilde M_{\mathrm{kraus},H}$ Hermitian positive semidefinite and $\tilde M_{\mathrm{kraus},0}$ a partial isometry; the positive factor is the outcome's strength, the partial isometry is its backaction, and the effect is the squared strength conjugated by the backaction, $\tilde{\mathcal E}_{\Psi}=\tilde M_{\mathrm{kraus}}^{*}\tilde M_{\mathrm{kraus}}=\tilde M_{\mathrm{kraus},0}^{*}\tilde M_{\mathrm{kraus},H}^{2}\tilde M_{\mathrm{kraus},0}$. Three consequences follow. A one-Kraus channel is trace preserving exactly when the strength is trivial, $\tilde M_{\mathrm{kraus},H}=e_t$, and then the channel is $*$-unitary. A sharp outcome has a rank-one amplitude on the null cone, where the strength is a scaled projection and the backaction a partial isometry. And the reversibility of a channel is its Kraus rank, not the strength of any single term: a channel can be a sum of terms whose individual polar factors are $*$-unitary and still be irreversible.

The distinction between one Kraus operator, an instrument, and a channel is the source of most misreadings: the polar decomposition is always available for one operator, while the one-Kraus unitarity and the multi-Kraus irreversibility are statements about a channel.

This article is the pilot of the polar programme in the physics corpus: the object is the Kraus operator of a measurement outcome, the examples are two-dimensional, and every numerical claim is checked in the matrix model. It does not derive the measurement postulate or classify channels; it tests the grammar of the generalized Cartan pairing on the object where that grammar has the most work to do.

The conventions are those of the companion articles. The algebra is $\mathbb{B}=\mathbb{C}\otimes_{\mathbb{R}}\mathbb{H}$ with basis $e_t,\vec e_x,\vec e_y,\vec e_z$, $\vec e_\eta^{2}=-e_t$, and central scalar imaginary $i$. The Hermitian conjugation is ${}^{*}$, the Hermitian subspace is $\mathbb{B}_{\Psi}$, the anti-Hermitian subspace is $\mathbb{B}_{\Phi}$. A state is $\tilde{Ϛ}_{\Psi}=\tfrac12(e_t+i\vec s_{\Psi})$ with $\vec s_{\Psi}\in\mathbb{R}^{3}$ and $|\vec s_{\Psi}|\leq1$. The positive factor of the polar decomposition is written $\tilde M_{\mathrm{kraus},H}$ and the partial isometry $\tilde M_{\mathrm{kraus},0}$; they are the $\tilde M_{\mathrm{kraus},H}$ and $\tilde M_{\mathrm{kraus},0}$ of the generalized Cartan pairing, and the subscripts keep them apart from the Hamiltonian $\tilde{H}_{\Psi,R}$ and the $*$-unitary element $\tilde M_{\mathrm{kraus},0}$ of a channel. The doubled scalar part is $2\,\mathrm{Sc}\tilde Q=2\operatorname{Sc}\tilde Q=2Q_t$, read at normalisation.

The Results It Uses

result established in used here for
the generalized Cartan pairing on all of $\mathbb{B}$ The Polar Decomposition of a Biquaternion, §The Isometry Factor $\tilde M_{\mathrm{kraus}}=\tilde M_{\mathrm{kraus},H}\tilde M_{\mathrm{kraus},0}$, $\tilde M_{\mathrm{kraus},H}=\sqrt{\tilde M_{\mathrm{kraus}}\tilde M_{\mathrm{kraus}}^{*}}$, $\tilde M_{\mathrm{kraus},0}=\tilde M_{\mathrm{kraus},H}^{+}\tilde M_{\mathrm{kraus}}$
the Cartan pairing off the null cone The Cartan Hermitian Pairing of the Polar Decomposition of Biquaternions, §The Statement the invertible case, where $\tilde M_{\mathrm{kraus},0}$ is $*$-unitary
the rank-one Hermitian idempotents Idempotents, §The Classification of the Idempotents the sharp outcome $\tilde M_{\mathrm{kraus}}=\tilde\Pi$ and the Bloch form of its projections
the zero-divisor cone N-Regularity and N-Degeneracy in the Biquaternion Algebra, §The Zero Divisor Set the boundary where the amplitude loses rank
the Kraus representation and the channel dichotomy Quantum Channels and the Reversible/Irreversible Dichotomy, §The Kraus Representation the instrument and channel context
the effect as a Kraus square POVMs and the Positive Cone in Biquaternionic Form, §The post-measurement state $\tilde{\mathcal E}_{\Psi}=\tilde M_{\mathrm{kraus}}^{*}\tilde M_{\mathrm{kraus}}$ and the Born weight

The Object and the Three Levels

One Operator, One Outcome, One Channel

An instrument is a family of completely positive maps indexed by an outcome. In the algebra it is a family of Kraus operators $\tilde M_{\mathrm{kraus},y}$ with

$$ \tilde{\mathcal E}_{\Psi,y}=\tilde M_{\mathrm{kraus},y}^{*}\tilde M_{\mathrm{kraus},y},\qquad \sum_y\tilde{\mathcal E}_{\Psi,y}=e_t,\qquad p_y=2\,\mathrm{Sc}\!\left(\tilde{Ϛ}_{\Psi}\,\tilde{\mathcal E}_{\Psi,y}\right),\qquad \tilde{Ϛ}_{\Psi,y}=\frac{\tilde M_{\mathrm{kraus},y}\tilde{Ϛ}_{\Psi}\tilde M_{\mathrm{kraus},y}^{*}}{p_y}. $$

Dropping the outcome label gives the channel

$$ Ξ(\tilde{Ϛ}_{\Psi})=\sum_y\tilde M_{\mathrm{kraus},y}\tilde{Ϛ}_{\Psi}\tilde M_{\mathrm{kraus},y}^{*}. $$

A single Kraus operator is a channel only when it is the whole instrument, that is, when $\tilde M_{\mathrm{kraus}}^{*}\tilde M_{\mathrm{kraus}}=e_t$. Otherwise it is one outcome among several, and its effect is not the identity.

The polar decomposition is a statement about $\tilde M_{\mathrm{kraus}}$, not about the channel. It exists for every $\tilde M_{\mathrm{kraus}}\in\mathbb{B}$, including $\tilde M_{\mathrm{kraus}}=0$. It is unique once the initial and final projections are fixed, and it does not require the trace-preservation condition. The channel statements enter only when the Kraus operators are summed or when the one-term condition is imposed.

The Three Statements That Must Not Be Confused

statement object condition content
polar form one Kraus operator none $\tilde M_{\mathrm{kraus}}=\tilde M_{\mathrm{kraus},H}\tilde M_{\mathrm{kraus},0}$
one-Kraus channel the map $\tilde{Ϛ}_{\Psi}\mapsto\tilde M_{\mathrm{kraus}}\tilde{Ϛ}_{\Psi}\tilde M_{\mathrm{kraus}}^{*}$ $\tilde M_{\mathrm{kraus}}^{*}\tilde M_{\mathrm{kraus}}=e_t$ $\tilde M_{\mathrm{kraus}}$ is $*$-unitary, $\tilde M_{\mathrm{kraus},H}=e_t$
irreversible channel the sum $\sum_y\tilde M_{\mathrm{kraus},y}\cdot\tilde M_{\mathrm{kraus},y}^{*}$ minimal Kraus rank $\geq2$ no single $*$-unitary represents the map

The three rows are independent. A single Kraus operator can have a non-trivial $\tilde M_{\mathrm{kraus},H}$ and still be a legitimate outcome of an instrument. A single Kraus operator can have a trivial $\tilde M_{\mathrm{kraus},H}$ and belong to an irreversible channel. A channel can be a sum of two terms whose individual $\tilde M_{\mathrm{kraus},H}$ are scalar and whose individual $\tilde M_{\mathrm{kraus},0}$ are $*$-unitary, and still be irreversible.

The Polar Form of a Kraus Operator

The Factorization

The generalized Cartan pairing, read with $\tilde Q=\tilde M_{\mathrm{kraus}}$, gives the following. For every $\tilde M_{\mathrm{kraus}}\in\mathbb{B}$ there is exactly one pair $(\tilde M_{\mathrm{kraus},H},\tilde M_{\mathrm{kraus},0})$ with

  1. $\tilde M_{\mathrm{kraus},H}$ Hermitian positive semidefinite and $\operatorname{ran}\tilde M_{\mathrm{kraus},H}=\operatorname{ran}\tilde M_{\mathrm{kraus}}$;
  2. $\tilde M_{\mathrm{kraus},0}$ a partial isometry, $\tilde M_{\mathrm{kraus},0}\tilde M_{\mathrm{kraus},0}^{*}\tilde M_{\mathrm{kraus},0}=\tilde M_{\mathrm{kraus},0}$, with $\tilde M_{\mathrm{kraus},0}^{*}\tilde M_{\mathrm{kraus},0}$ the projection onto $\operatorname{ran}\tilde M_{\mathrm{kraus}}^{*}$ and $\tilde M_{\mathrm{kraus},0}\tilde M_{\mathrm{kraus},0}^{*}$ the projection onto $\operatorname{ran}\tilde M_{\mathrm{kraus}}$;
  3. $\tilde M_{\mathrm{kraus}}=\tilde M_{\mathrm{kraus},H}\tilde M_{\mathrm{kraus},0}$.

It is

$$ \boxed{\;\tilde M_{\mathrm{kraus},H}=\sqrt{\tilde M_{\mathrm{kraus}}\tilde M_{\mathrm{kraus}}^{*}},\qquad \tilde M_{\mathrm{kraus},0}=\tilde M_{\mathrm{kraus},H}^{+}\tilde M_{\mathrm{kraus}}\;}, $$

with $\tilde M_{\mathrm{kraus},H}^{+}$ the Moore–Penrose inverse. This is the generalized polar decomposition of the mathematical article; it is cited, not re-proved.

Strength and Backaction

In the singular value form, with at most two terms because $\mathbb{B}\cong M_2(\mathbb{C})$,

$$ \tilde M_{\mathrm{kraus}}=\sum_i\sigma_i\,u_i v_i^{*},\qquad \tilde M_{\mathrm{kraus},H}=\sum_i\sigma_i\,u_i u_i^{*},\qquad \tilde M_{\mathrm{kraus},0}=\sum_i u_i v_i^{*}, $$

with $\sigma_i\geq0$, and $u_i,v_i$ orthonormal in the matrix model. The two readings are:

  • strength: the unordered pair $\{\sigma_1,\sigma_2\}$, the eigenvalues of $\tilde M_{\mathrm{kraus},H}$. It is basis-independent; it is the list of singular values of $\tilde M_{\mathrm{kraus}}$.
  • backaction: the partial isometry $\tilde M_{\mathrm{kraus},0}$, which carries the input direction $v_i$ to the output direction $u_i$. Off the cone it is $*$-unitary; on the cone it has rank one.

The strength is what the outcome probability can see: the effect has eigenvalues $\sigma_i^{2}$. The backaction is what the probability discards: two amplitudes with the same singular values but different $\tilde M_{\mathrm{kraus},0}$ have the same effect and different post-measurement states.

The domains are fixed once and for all:

statement domain factors
generalized polar form every $\tilde M_{\mathrm{kraus}}\in\mathbb{B}$, including $0$ $\tilde M_{\mathrm{kraus},H}\geq0$, $\tilde M_{\mathrm{kraus},0}$ a partial isometry
Cartan pairing off the cone $\langle\tilde M_{\mathrm{kraus}}\rangle_{\natural}\neq0$ $\tilde M_{\mathrm{kraus},H}>0$, $\tilde M_{\mathrm{kraus},0}$ $*$-unitary
one-Kraus channel $\tilde M_{\mathrm{kraus}}^{*}\tilde M_{\mathrm{kraus}}=e_t$ $\tilde M_{\mathrm{kraus},H}=e_t$, $\tilde M_{\mathrm{kraus},0}=\tilde M_{\mathrm{kraus}}$ $*$-unitary
null case $\langle\tilde M_{\mathrm{kraus}}\rangle_{\natural}=0$, $\tilde M_{\mathrm{kraus}}\neq0$ $\operatorname{rank}\tilde M_{\mathrm{kraus},H}=1$, $\tilde M_{\mathrm{kraus},0}$ rank one

The strength decides the case of the operator, not the case of the channel; the four cases are collected in §The Three Strata.

Trace Preservation Is the Triviality of the Strength

The One-Kraus Channel

Let $\tilde M_{\mathrm{kraus}}$ be the single Kraus operator of a channel. Trace preservation is

$$ \tilde M_{\mathrm{kraus}}^{*}\tilde M_{\mathrm{kraus}}=e_t . $$

In $\mathbb{B}\cong M_2(\mathbb{C})$ this makes $\tilde M_{\mathrm{kraus}}$ invertible, with inverse $\tilde M_{\mathrm{kraus}}^{*}$; then $\tilde M_{\mathrm{kraus}}\tilde M_{\mathrm{kraus}}^{*}=e_t$ as well. Hence

$$ \tilde M_{\mathrm{kraus},H}=\sqrt{\tilde M_{\mathrm{kraus}}\tilde M_{\mathrm{kraus}}^{*}}=e_t,\qquad \tilde M_{\mathrm{kraus},0}=\tilde M_{\mathrm{kraus}}, $$

and $\tilde M_{\mathrm{kraus}}$ is $*$-unitary. Conversely, if $\tilde M_{\mathrm{kraus},H}=e_t$ and $\tilde M_{\mathrm{kraus},0}$ is $*$-unitary, then $\tilde M_{\mathrm{kraus}}=\tilde M_{\mathrm{kraus},0}$ and $\tilde M_{\mathrm{kraus}}^{*}\tilde M_{\mathrm{kraus}}=e_t$. Therefore:

A one-Kraus channel is trace preserving if and only if its positive factor is trivial; the channel is then conjugation by a $*$-unitary biquaternion.

This is the reversible case of Quantum Channels and the Reversible/Irreversible Dichotomy, read in the polar language. The Hamiltonian $\tilde{H}_{\Psi,R}\in\mathbb{B}_{\Psi}$ generates the $*$-unitary factor through $\tilde M_{\mathrm{kraus},0}=\exp(-i\theta_{\mathrm{evo}})$, and the generator $-\frac{i\,H_{R}}{\hbar}$ lies in $\mathbb{B}_{\Phi}$. The polar factor $\tilde M_{\mathrm{kraus},0}$ is the evolution itself; its logarithm is the material generator. The condition $\tilde M_{\mathrm{kraus},H}=e_t$ says that the evolution attenuates nothing.

The angle is $\theta_{\mathrm{evo}}$, the scalar part of $\tilde H_\Psi$ times the interval over $\hbar\,c$: the generator is anti-Hermitian, so its scalar part is purely imaginary and the angle is real.

What Trace Preservation Does Not Say

Trace preservation of a one-Kraus channel is not the same as trace preservation of an instrument. An instrument satisfies $\sum_y\tilde M_{\mathrm{kraus},y}^{*}\tilde M_{\mathrm{kraus},y}=e_t$; a single term need not. A term with $\tilde M_{\mathrm{kraus},H}\neq e_t$ is a legitimate outcome, and its effect is not the identity. The one-Kraus statement applies only when there is one term and no other outcome. The dephasing channel of §Channels and Kraus Rank is the witness: one Kraus operator is a scalar multiple of the identity, with $\tilde M_{\mathrm{kraus},0}=e_t$, and the channel is still irreversible because there is a second term.

The Four Factors of the Amplitude

The Word Off the Cone

Off the null cone the two polar factors refine into the four factors of the complete polar decomposition. With $\tilde U_{\Psi,\mathrm b}=\exp\bigl(i\tfrac{\theta_{\mathrm{rpd},w}}{2}\vec u_{\mathrm b}\bigr)$ a positive Hermitian boost and $\tilde u_{\mathrm r}=\exp\bigl(\tfrac{\theta_{\mathrm{axr}}}{2}\vec u_{\mathrm r}\bigr)$ a unit real quaternion,

$$ \tilde M_{\mathrm{kraus}}=\varrho\,e^{i\theta_{\mathrm{pha}}}\,\tilde U_{\Psi,\mathrm b}\,\tilde u_{\mathrm r}, \qquad \tilde M_{\mathrm{kraus},H}=\varrho\tilde U_{\Psi,\mathrm b}, \qquad \tilde M_{\mathrm{kraus},0}=e^{i\theta_{\mathrm{pha}}}\tilde u_{\mathrm r}. $$

Here $\theta_{\mathrm{pha}}$ is the phase of the polar form: it is the argument of the central complex scalar, a real angle taken modulo a full turn, and the exponential that carries it is $\exp(i\theta_{\mathrm{pha}})$.

The grouping is the Cartan one: the scale sits with the boost and the phase with the rotor. The four factors are a change of coordinates on the invertible elements, not a further hypothesis, and they carry the eight real parameters of $\tilde M_{\mathrm{kraus}}$.

factor symbol range parameters factor of the pair
scale $\varrho$ $\varrho>0$ $1$ $\tilde M_{\mathrm{kraus},H}$
phase $e^{i\theta_{\mathrm{pha}}}$ $\theta_{\mathrm{pha}}\in[0,2\pi)$ $1$ $\tilde M_{\mathrm{kraus},0}$
boost $\tilde U_{\Psi,\mathrm b}$ $\theta_{\mathrm{rpd},w}\ge0$, unit $\vec u_{\mathrm b}$ $3$ $\tilde M_{\mathrm{kraus},H}$
rotor $\tilde u_{\mathrm r}$ $\theta_{\mathrm{axr}}\in[0,2\pi)$, unit $\vec u_{\mathrm r}$ $3$ $\tilde M_{\mathrm{kraus},0}$

The Effect in the Word

The effect is the square of the positive factor, carried to the initial frame by the backaction:

$$ \tilde{\mathcal E}_{\Psi}=\tilde M_{\mathrm{kraus}}^{*}\tilde M_{\mathrm{kraus}} =\tilde M_{\mathrm{kraus},0}^{*}\tilde M_{\mathrm{kraus},H}^{2}\tilde M_{\mathrm{kraus},0} =\tilde u_{\mathrm r}^{-1}\,\varrho^{2}\tilde U_{\Psi,\mathrm b}^{2}\,\tilde u_{\mathrm r}. $$

The phase cancels because it is central. Since $\tilde U_{\Psi,\mathrm b}^{2}=\exp(i\theta_{\mathrm{rpd},w}\vec u_{\mathrm b})$ has eigenvalues $e^{\pm\theta_{\mathrm{rpd},w}}$, the effect has eigenvalues

$$ \varrho^{2}e^{+\theta_{\mathrm{rpd},w}}\quad\text{and}\quad \varrho^{2}e^{-\theta_{\mathrm{rpd},w}}. $$

Its eigenvector for the larger eigenvalue is $\tilde u_{\mathrm r}^{-1}\vec u_{\mathrm b}$, the boost axis turned back by the rotor, for $\theta_{\mathrm{rpd},w}>0$. Generically, of the eight parameters the effect keeps four — the scale, the rapidity and the composite direction $\tilde u_{\mathrm r}^{-1}\vec u_{\mathrm b}$ — and loses the phase, the rotation angle and the split between the two axes. This is the quantitative form of the statement that the probability sees the strength and not the backaction.

The Boost as a Bias

The rotor and the boost are not the same kind of factor. On a state $\tilde{Ϛ}_{\Psi}=\tfrac12(e_t+i\vec s_{\Psi})$ the rotor acts by conjugation, which fixes the scalar part and rotates the Bloch vector by $\theta_{\mathrm{axr}}$ about $\vec u_{\mathrm r}$: the trace is one, the radius is unchanged, and the state stays on its sphere.

The boost changes the trace, mixes the scalar and parallel parts, and leaves the transverse part alone. Writing $\tilde{Ϛ}_{\Psi}=a e_t+i\vec b$ with $a=\tfrac12$ and $\vec b=\tfrac12\vec s_{\Psi}$,

$$ a_2=a\cosh\theta_{\mathrm{rpd},w}+b_\parallel\sinh\theta_{\mathrm{rpd},w}, \qquad b_{\parallel,2}=b_\parallel\cosh\theta_{\mathrm{rpd},w}+a\sinh\theta_{\mathrm{rpd},w}, \qquad \vec b_{\perp,2}=\vec b_\perp . $$

Normalising by $a_2$ gives the action on the Bloch ball,

$$ r'_\parallel=\frac{s_\parallel+\tanh\theta_{\mathrm{rpd},w}}{1+s_\parallel\tanh\theta_{\mathrm{rpd},w}}, \qquad \vec s_{\Psi,\perp}'=\frac{\vec s_{\Psi,\perp}}{\cosh\theta_{\mathrm{rpd},w}+s_\parallel\sinh\theta_{\mathrm{rpd},w}}, $$

with $s_\parallel=\vec s_{\Psi}\cdot\vec u_{\mathrm b}$. The poles $\pm\vec u_{\mathrm b}$ are fixed, the boundary sphere is mapped to itself, and every interior point moves toward $+\vec u_{\mathrm b}$. This is velocity addition along the boost axis, with strength $\tanh\theta_{\mathrm{rpd},w}$, and it is not a rotation of the ball.

The Contraction Bound

An outcome is admissible when its effect lies below the identity. The eigenvalues above make this one inequality:

$$ \boxed{\;\tilde{\mathcal E}_{\Psi}\le e_t \iff \varrho^{2}e^{\theta_{\mathrm{rpd},w}}\le1\;} $$

for $\theta_{\mathrm{rpd},w}\ge0$, which forces $\varrho\le1$. The scale and the rapidity trade against each other, and a strong boost needs a small scale.

The worked example of §A Worked Unsharp Outcome saturates the bound. Its strength has eigenvalues $0.6$ and $1$, so $\varrho=\sqrt{0.6}=0.7746$ and $\theta_{\mathrm{rpd},w}=\ln(1/0.6)=0.5108$; then $\varrho^{2}e^{\theta_{\mathrm{rpd},w}}=0.6\times1.6667=1$ exactly, and the effect has eigenvalues $0.36$ and $1$. The complement of the effect is rank one, as that example notes, and that is the same statement.

Effects as Squares

The Effect Is the Conjugated Square

The effect of the outcome is

$$ \boxed{\;\tilde{\mathcal E}_{\Psi}=\tilde M_{\mathrm{kraus}}^{*}\tilde M_{\mathrm{kraus}}=\tilde M_{\mathrm{kraus},0}^{*}\tilde M_{\mathrm{kraus},H}^{2}\tilde M_{\mathrm{kraus},0}\;}. $$

The eigenvalues of $\tilde{\mathcal E}_{\Psi}$ are $\sigma_i^{2}$, the squares of the singular values of $\tilde M_{\mathrm{kraus}}$ and the eigenvalues of $\tilde M_{\mathrm{kraus},H}^{2}$. The effect is positive, and $\tilde{\mathcal E}_{\Psi}\leq e_t$ exactly when every $\sigma_i\leq1$; for an instrument the resolution $\sum_y\tilde{\mathcal E}_{\Psi,y}=e_t$ forces each $\sigma_i\leq1$.

The formula is not $\tilde{\mathcal E}_{\Psi}=\tilde M_{\mathrm{kraus},H}^{2}$ in general. The two agree only when $\tilde M_{\mathrm{kraus},0}$ commutes with $\tilde M_{\mathrm{kraus},H}$, and in particular when $\tilde M_{\mathrm{kraus},0}$ is a function of $\tilde M_{\mathrm{kraus},H}$. On the cone they need not agree at all: the effect and the squared strength can sit on different lines, with $\tilde M_{\mathrm{kraus},0}$ carrying one to the other.

The Effect Does Not Determine the Split

The effect determines the singular values, not the factorization. If $\tilde U_{\mathrm{gau}}$ is $*$-unitary and $\tilde M_{\mathrm{kraus}}'=\tilde U_{\mathrm{gau}}\tilde M_{\mathrm{kraus}}$, then

$$ \tilde M_{\mathrm{kraus}}'^{*}\tilde M_{\mathrm{kraus}}'=\tilde M_{\mathrm{kraus}}^{*}\tilde M_{\mathrm{kraus}}=\tilde{\mathcal E}_{\Psi}, \qquad \tilde M_{\mathrm{kraus},H}'=\tilde U_{\mathrm{gau}}\tilde M_{\mathrm{kraus},H}\tilde U_{\mathrm{gau}}^{*}, \qquad \tilde M_{\mathrm{kraus},0}'=\tilde U_{\mathrm{gau}}\tilde M_{\mathrm{kraus},0}. $$

The same effect therefore admits different strengths and different backactions. The canonical choice $\tilde M_{\mathrm{kraus}}=\sqrt{\tilde{\mathcal E}_{\Psi}}$ gives $\tilde M_{\mathrm{kraus},H}=\sqrt{\tilde{\mathcal E}_{\Psi}}$ and $\tilde M_{\mathrm{kraus},0}$ the projection onto $\operatorname{ran}\tilde{\mathcal E}_{\Psi}$; a physical Kraus operator need not be that choice. The polar decomposition is a decomposition of the amplitude, not a decomposition of the effect. This is the precise sense in which the effect is the probability functional and the amplitude carries more.

A Null Example Where the Two Differ

Take the non-Hermitian null element

$$ \tilde M_{\mathrm{kraus}}=i\vec e_x-\vec e_y,\qquad \langle\tilde M_{\mathrm{kraus}}\rangle_{\natural}=i^{2}+(-1)^{2}=0. $$

Its singular values are $2$ and $0$. The generalized pairing gives

$$ \tilde M_{\mathrm{kraus},H}=e_t+i\vec e_z=2\tilde\Pi_{\vec e_z},\qquad \tilde M_{\mathrm{kraus},0}=\tfrac{i}{2}\vec e_x-\tfrac12\vec e_y=\tfrac12\tilde M_{\mathrm{kraus}} . $$

The effect and the squared strength are

$$ \boxed{\;\tilde{\mathcal E}_{\Psi}=\tilde M_{\mathrm{kraus}}^{*}\tilde M_{\mathrm{kraus}}=4\tilde\Pi_{-\vec e_z}=2e_t-2i\vec e_z, \qquad \tilde M_{\mathrm{kraus},H}^{2}=4\tilde\Pi_{\vec e_z}=2e_t+2i\vec e_z\;} . $$

They are different rank-one elements, on opposite Bloch points. The partial isometry carries the initial line $\tilde\Pi_{-\vec e_z}$ to the final line $\tilde\Pi_{\vec e_z}$, and the effect sees the initial line while the squared strength sees the final one. The example is exact; the identities $\tilde M_{\mathrm{kraus}}=\tilde M_{\mathrm{kraus},H}\tilde M_{\mathrm{kraus},0}$ and $\tilde{\mathcal E}_{\Psi}=\tilde M_{\mathrm{kraus},0}^{*}\tilde M_{\mathrm{kraus},H}^{2}\tilde M_{\mathrm{kraus},0}$ hold with no residual.

The Three Strata

The four cases of the amplitude are the four rows of the table. The rank is the matrix rank of $\mathsf M_2(\tilde M_{\mathrm{kraus}})$; the projections are $\tilde\Pi_{\mathrm i}=\tilde M_{\mathrm{kraus},0}^{*}\tilde M_{\mathrm{kraus},0}$ and $\tilde\Pi_{\mathrm f}=\tilde M_{\mathrm{kraus},0}\tilde M_{\mathrm{kraus},0}^{*}$.

stratum $\operatorname{rank}\tilde M_{\mathrm{kraus}}$ $\tilde M_{\mathrm{kraus},H}$ $\tilde M_{\mathrm{kraus},0}$ effect $\tilde{\mathcal E}_{\Psi}$ example
$*$-unitary $2$ $e_t$ $\tilde M_{\mathrm{kraus}}$, $*$-unitary $e_t$ reversible one-Kraus channel
invertible, not $*$-unitary $2$ positive definite, $\neq e_t$ $*$-unitary, $\neq\tilde M_{\mathrm{kraus}}$ positive definite, $ unsharp outcome
null, rank one $1$ $\sigma\tilde\Pi_{\mathrm f}$, $\sigma>0$ rank-one partial isometry $\sigma^{2}\tilde\Pi_{\mathrm i}$ sharp ($\sigma=1$) or soft ($\sigma<1$) outcome
zero $0$ $0$ $0$ $0$ null outcome

Sharp Measurement on the Null Cone

Take the rank-one Hermitian idempotent

$$ \tilde M_{\mathrm{kraus}}=\tilde\Pi=\tfrac12\left(e_t+i\vec u\right),\qquad \vec u^{2}=-e_t. $$

It satisfies $\tilde\Pi^{2}=\tilde\Pi=\tilde\Pi^{*}$ and $\langle\tilde\Pi\rangle_{\natural}=0$, so it is a zero divisor. Its polar factors are $\tilde M_{\mathrm{kraus},H}=\tilde M_{\mathrm{kraus},0}=\tilde\Pi$, because $\tilde\Pi\tilde\Pi^{*}=\tilde\Pi$ and $\tilde\Pi^{+}=\tilde\Pi$. The effect is $\tilde{\mathcal E}_{\Psi}=\tilde\Pi$, and the two projections of the partial isometry are both $\tilde\Pi$. The outcome probability for a state $\tilde{Ϛ}_{\Psi}$ is $p=2\,\mathrm{Sc}(\tilde{Ϛ}_{\Psi}\tilde\Pi)$, and the post-measurement state is $\tilde\Pi$.

The sharp outcome therefore sits exactly on the null cone, and its amplitude is a partial isometry, not a $*$-unitary. The "collapse" is the rank drop: the amplitude has rank one, the effect has rank one, and the post-measurement state is a pure state. The mathematical pairing that covers it is the generalized Cartan pairing; the Cartan pairing on $\langle\tilde M_{\mathrm{kraus}}\rangle_{\natural}\neq0$ does not.

A projective measurement of a two-level system is the pair $\tilde\Pi_1,\tilde\Pi_2$ with $\tilde\Pi_1+\tilde\Pi_2=e_t$. Both Kraus operators are null; both effects are idempotent; the instrument is sharp. The resolution of the identity is the statement $\tilde\Pi_1+\tilde\Pi_2=e_t$, not a statement about either factor alone.

Weak Measurement and Backaction

The Invertible Outcome

An outcome is unsharp when $\tilde M_{\mathrm{kraus}}$ is invertible. Then $\tilde M_{\mathrm{kraus},H}$ is positive definite, $\tilde M_{\mathrm{kraus},0}$ is $*$-unitary, and the effect is positive definite. The post-measurement state is

$$ \tilde{Ϛ}_{\Psi,y}=\frac{\tilde M_{\mathrm{kraus},H}\tilde M_{\mathrm{kraus},0}\tilde{Ϛ}_{\Psi}\tilde M_{\mathrm{kraus},0}^{*}\tilde M_{\mathrm{kraus},H}}{p_y}. $$

The order is fixed by the word: the $*$-unitary factor acts first, then the positive filter. The backaction is the rotation $\tilde M_{\mathrm{kraus},0}$; the filter is the strength $\tilde M_{\mathrm{kraus},H}$ in the output basis. A weak measurement is the regime in which the $\sigma_i$ are close to $1$; the outcome then disturbs the state only slightly, and the effect is close to $e_t$.

A Worked Unsharp Outcome

Take the strength $\tilde{H}_{\Psi,0}=\tfrac{1+\eta}{2}e_t+\tfrac{\eta-1}{2}i\vec e_z=0.8e_t-0.2i\vec e_z$ with $\eta=0.6$ and eigenvalues $0.6$ and $1$, and the $*$-unitary backaction $\tilde Q_{\mathrm{ba}}=\cos\tfrac{\theta_{\mathrm{axr}}}{2}e_t+\sin\tfrac{\theta_{\mathrm{axr}}}{2}\vec e_x=0.866025404e_t+0.5\vec e_x$ with $\theta_{\mathrm{axr}}=\pi/3$. Set $\tilde M_{\mathrm{kraus}}=\tilde Q_{\mathrm{ba}}\tilde{H}_{\Psi,0}$. The polar factors are

$$ \tilde M_{\mathrm{kraus},H}=\tilde Q_{\mathrm{ba}}\tilde{H}_{\Psi,0}\tilde Q_{\mathrm{ba}}^{*}=0.8e_t+0.173205081i\vec e_y-0.1i\vec e_z,\qquad \tilde M_{\mathrm{kraus},0}=\tilde Q_{\mathrm{ba}}, $$

and the amplitude is $\tilde M_{\mathrm{kraus}}=0.692820323e_t+0.4\vec e_x+0.1i\vec e_y-0.173205081i\vec e_z$. The effect is

$$ \boxed{\;\tilde{\mathcal E}_{\Psi}=\tilde M_{\mathrm{kraus}}^{*}\tilde M_{\mathrm{kraus}}=\tilde{H}_{\Psi,0}^{2}=0.68e_t-0.32i\vec e_z=0.36\tilde\Pi_{\vec e_z}+\tilde\Pi_{-\vec e_z}\;}, $$

with eigenvalues $0.36$ and $1$. The outcome is unsharp: the effect is full rank, no direction is annihilated, and the backaction is a rotation by $60^{\circ}$ about $\vec e_x$. The strength is the pair $\{0.6,1\}$; the effect sees its squares. The complement $e_t-\tilde{\mathcal E}_{\Psi}=0.64\tilde\Pi_{\vec e_z}$ is rank one, with Kraus operator $0.8\tilde\Pi_{\vec e_z}$, which is null. The two outcomes resolve the identity: one is unsharp and invertible, the other is rank one and on the cone. The example shows the boundary in a single instrument.

Channels and Kraus Rank

The Sum of the Terms

A channel is a sum of terms, and the polar decomposition applies to each term separately:

$$ Ξ(\tilde{Ϛ}_{\Psi})=\sum_l\tilde M_{\mathrm{kraus},l}\tilde{Ϛ}_{\Psi}\tilde M_{\mathrm{kraus},l}^{*} =\sum_l\tilde H_{\Psi,\mathrm{pos},l}\tilde U_{\mathrm{p},l}\tilde{Ϛ}_{\Psi}\tilde U_{\mathrm{p},l}^{*}\tilde H_{\Psi,\mathrm{pos},l}. $$

Trace preservation is

$$ \sum_l\tilde U_{\mathrm{p},l}^{*}\tilde H_{\Psi,\mathrm{pos},l}^{2}\tilde U_{\mathrm{p},l}=e_t. $$

The polar decomposition of the sum is not the sum of the polar decompositions. The object decomposed here is each Kraus operator; the channel is a completely positive map, and its own polar decomposition as an operator on the Hilbert–Schmidt space is a different statement, not made here.

The Dichotomy Is the Minimal Number of Terms

A channel is reversible exactly when it has a one-term Kraus representation, and then the term is $*$-unitary. It is irreversible exactly when its minimal Kraus rank is at least two. The value of one $\tilde M_{\mathrm{kraus},H}$ does not decide the channel: a term can have $\tilde M_{\mathrm{kraus},H}=e_t$ and the channel can still be irreversible, and a term can have $\tilde M_{\mathrm{kraus},H}\neq e_t$ and the channel can still be reversible if the representation is non-minimal and the map has a one-term form.

Dephasing as the Counterexample

The dephasing channel of Quantum Channels and the Reversible/Irreversible Dichotomy leaves the populations and scales the coherences by $1-p$. In the pointer basis it is

$$ Ξ_p(\tilde{Ϛ}_{\Psi})=(1-p)\tilde{Ϛ}_{\Psi}+p\left(\tilde\Pi_1\tilde{Ϛ}_{\Psi}\tilde\Pi_1+\tilde\Pi_2\tilde{Ϛ}_{\Psi}\tilde\Pi_2\right). $$

The companion article displays the three-term representation $\tilde M_{\mathrm{kraus},0}=\sqrt{1-p}e_t$, $\tilde M_{\mathrm{kraus},1,2}=\sqrt p\tilde\Pi_{1,2}$. That representation is not minimal: because $e_t=\tilde\Pi_1+\tilde\Pi_2$, the same channel has the two-term form

$$ \tilde M_{\mathrm{kraus},0}=\sqrt{1-\tfrac p2}\,e_t,\qquad \tilde M_{\mathrm{kraus},1}=\sqrt{\tfrac p2}\,Z,\qquad Z=i\vec e_z, $$

with $Z$ $*$-unitary and $Z^{2}=e_t$. For $00$, because its minimal Kraus rank is two. At $p=0$ the second term vanishes and the channel is the identity; at $p=1$ the two terms are $e_t/\sqrt2$ and $Z/\sqrt2$ and the channel is fully dephasing.

The example states the article's negative thesis. The reversibility of a channel is not the triviality of a single strength and not the unitarity of a single backaction. It is the minimal number of terms in the sum.

Physical Interpretation

The polar form separates two questions that the outcome probability answers together. The strength answers how much the outcome responds to each direction; the backaction answers which way the state is carried when the outcome occurs. The probability is the trace of the effect, so it sees only the squares of the strengths. The post-measurement state sees the backaction as well. Two amplitudes with the same effect but different $\tilde M_{\mathrm{kraus},0}$ are indistinguishable in their statistics and distinguishable in their aftermath; the polar form is what makes the difference explicit.

The $*$-unitary stratum is the case of no filter. The outcome rotates the state and does not weight it; if it is the whole channel, the process is reversible and its generator is a Hamiltonian in $\mathbb{B}_{\Phi}$. The invertible stratum is the unsharp outcome: the state is rotated and then weighted, no direction is annihilated, and the effect lies strictly inside the effect body. The null stratum is the sharp or soft rank-one outcome: the amplitude projects onto a line, the effect is rank one, and the post-measurement state is pure. The three strata are the three physical regimes of a measurement outcome, read from the rank of the amplitude.

The null cone is the measurement boundary in a precise but limited sense. It is the locus where the amplitude loses rank, the boundary of the positive cone, and the boundary between the invertible and the rank-one outcomes. It is not a boundary in space or time, and it is not a dynamical condition that says when a measurement occurs. The algebra classifies the amplitude; it does not select it. The companion article The Measurement Problem in Algebraic Form owns that distinction, and this article does not improve on it.

The channel-level statement is the correction that the polar programme makes to a natural but false reading. It is tempting to say that a process is reversible when its positive factor is trivial, or that a process is a measurement when its backaction is a projector. Neither is a statement about a channel. A channel is a sum, and its reversibility is the minimal number of terms. The dephasing channel is the counterexample: two scalar strengths and two $*$-unitary backactions, and an irreversible map.

What This Does Not Claim

  • It does not prove the generalized polar decomposition. The existence, uniqueness and projection conditions are owned by the mathematical article and are cited.
  • It does not classify channels by the polar factor of one Kraus operator. The channel dichotomy is the minimal Kraus rank; the polar factor is a property of a term.
  • It does not identify $\tilde M_{\mathrm{kraus},H}$ with a physical field or $\tilde M_{\mathrm{kraus},0}$ with a force. The strength and backaction are readings of the amplitude, not new dynamics.
  • It does not claim that the null cone is a physical boundary in space or time. It is the singular locus of the algebra.
  • It does not resolve the measurement problem. It restates the classification of the acting element; the selection of the outcome and of the $*$-unitary-versus-idempotent case is not in the algebra.
  • It does not treat continuous POVMs, non-completely-positive maps, or the polar decomposition of a channel as a superoperator.

Summary

Main Results

item result
polar form $\tilde M_{\mathrm{kraus}}=\tilde M_{\mathrm{kraus},H}\tilde M_{\mathrm{kraus},0}$, $\tilde M_{\mathrm{kraus},H}=\sqrt{\tilde M_{\mathrm{kraus}}\tilde M_{\mathrm{kraus}}^{*}}$, $\tilde M_{\mathrm{kraus},0}=\tilde M_{\mathrm{kraus},H}^{+}\tilde M_{\mathrm{kraus}}$
strength the singular values $\sigma_i$, the eigenvalues of $\tilde M_{\mathrm{kraus},H}$
backaction the partial isometry $\tilde M_{\mathrm{kraus},0}$, $*$-unitary off the cone, rank one on it
effect $\tilde{\mathcal E}_{\Psi}=\tilde M_{\mathrm{kraus}}^{*}\tilde M_{\mathrm{kraus}}=\tilde M_{\mathrm{kraus},0}^{*}\tilde M_{\mathrm{kraus},H}^{2}\tilde M_{\mathrm{kraus},0}$, eigenvalues $\sigma_i^{2}$
four factors $\tilde M_{\mathrm{kraus}}=\varrho e^{i\theta_{\mathrm{pha}}}\tilde U_{\Psi,\mathrm b}\tilde u_{\mathrm r}$, $\tilde M_{\mathrm{kraus},H}=\varrho\tilde U_{\Psi,\mathrm b}$, $\tilde M_{\mathrm{kraus},0}=e^{i\theta_{\mathrm{pha}}}\tilde u_{\mathrm r}$
effect in the word eigenvalues $\varrho^{2}e^{\pm\theta_{\mathrm{rpd},w}}$; admissible iff $\varrho^{2}e^{\theta_{\mathrm{rpd},w}}\le1$
one-Kraus channel $\tilde M_{\mathrm{kraus}}^{*}\tilde M_{\mathrm{kraus}}=e_t$ iff $\tilde M_{\mathrm{kraus},H}=e_t$ and $\tilde M_{\mathrm{kraus},0}=\tilde M_{\mathrm{kraus}}$ $*$-unitary
sharp outcome $\tilde M_{\mathrm{kraus}}=\tilde\Pi$: $\tilde M_{\mathrm{kraus},H}=\tilde M_{\mathrm{kraus},0}=\tilde\Pi$, $\tilde{\mathcal E}_{\Psi}=\tilde\Pi$, on the null cone
soft rank-one outcome $\tilde M_{\mathrm{kraus}}=\sigma\tilde\Pi$: $\tilde M_{\mathrm{kraus},H}=\sigma\tilde\Pi$, $\tilde M_{\mathrm{kraus},0}=\tilde\Pi$, $\tilde{\mathcal E}_{\Psi}=\sigma^{2}\tilde\Pi$
unsharp outcome $\tilde M_{\mathrm{kraus}}$ invertible: $\tilde M_{\mathrm{kraus},H}>0$, $\tilde M_{\mathrm{kraus},0}$ $*$-unitary, $\tilde{\mathcal E}_{\Psi}>0$
channel $\sum_l\tilde U_{\mathrm{p},l}^{*}\tilde H_{\Psi,\mathrm{pos},l}^{2}\tilde U_{\mathrm{p},l}=e_t$; reversible iff minimal Kraus rank one
dephasing two scalar strengths and two $*$-unitary backactions, minimal Kraus rank two, irreversible for $p>0$

Summary of Notation

symbol meaning
$\mathbb{B}=\mathbb{C}\otimes_{\mathbb{R}}\mathbb{H}$ biquaternion algebra, $\cong M_2(\mathbb{C})$
$e_t,\vec e_x,\vec e_y,\vec e_z$ quaternion basis, $\vec e_\eta^{2}=-e_t$
$\tilde{Ϛ}_{\Psi}=\tfrac12(e_t+i\vec s_{\Psi})$ state, $\lvert\vec s_{\Psi}\rvert\leq1$
$\tilde M_{\mathrm{kraus}}$, $\tilde M_{\mathrm{kraus},y}$, $\tilde M_{\mathrm{kraus},l}$ Kraus operator, one outcome, one term of a channel
$\tilde M_{\mathrm{kraus},H}=\sqrt{\tilde M_{\mathrm{kraus}}\tilde M_{\mathrm{kraus}}^{*}}$ positive (strength) factor of the Kraus operator
$\tilde M_{\mathrm{kraus},0}=\tilde M_{\mathrm{kraus},H}^{+}\tilde M_{\mathrm{kraus}}$ partial-isometry (backaction) factor
$\tilde M_{\mathrm{kraus},H}^{+}$ Moore–Penrose inverse
$\tilde{\mathcal E}_{\Psi}=\tilde M_{\mathrm{kraus}}^{*}\tilde M_{\mathrm{kraus}}=\tilde M_{\mathrm{kraus},0}^{*}\tilde M_{\mathrm{kraus},H}^{2}\tilde M_{\mathrm{kraus},0}$ effect of the outcome
$\varrho,\theta_{\mathrm{pha}},\theta_{\mathrm{rpd},w},\theta_{\mathrm{axr}}$ scale, central phase, word rapidity, rotor angle
$\tilde U_{\Psi,\mathrm b},\tilde u_{\mathrm r}$ positive Hermitian boost, unit real quaternion rotor
$\tilde\Pi=\tfrac12(e_t+i\vec u)$ rank-one Hermitian idempotent, $\vec u$ a real unit vector
$\tilde\Pi_{\mathrm i}=\tilde M_{\mathrm{kraus},0}^{*}\tilde M_{\mathrm{kraus},0}$, $\tilde\Pi_{\mathrm f}=\tilde M_{\mathrm{kraus},0}\tilde M_{\mathrm{kraus},0}^{*}$ initial and final projections of $\tilde M_{\mathrm{kraus},0}$
$\sigma_i$ singular values of $\tilde M_{\mathrm{kraus}}$, eigenvalues of $\tilde M_{\mathrm{kraus},H}$
$Ξ(\tilde{Ϛ}_{\Psi})=\sum_l\tilde M_{\mathrm{kraus},l}\tilde{Ϛ}_{\Psi}\tilde M_{\mathrm{kraus},l}^{*}$ channel
$Z=i\vec e_z$ the $*$-unitary Pauli-$z$ element, $Z^{2}=e_t$
$2\,\mathrm{Sc}\tilde Q=2\operatorname{Sc}\tilde Q=2Q_t$ reduction convention
$\langle\tilde Q\rangle_{\natural}=\sum_\mu Q_\mu^{2}$ biquaternion norm

Further Reading