--- title: Cartan–Khaneja–Glaser Decomposition url: https://www.emergentmind.com/topics/cartan-khaneja-glaser-decomposition type: topic --- # Cartan–Khaneja–Glaser Decomposition The Cartan–Khaneja–Glaser decomposition, often written as a KAK or KHK decomposition, is a Lie-theoretic factorization of unitary operators that is specialized in quantum information to \(U \in SU(2^n)\) and related dynamical Lie groups. Its core statement is that, after choosing a Cartan involution and the associated symmetric-space splitting of the Lie algebra into compact and noncompact parts, every unitary can be written as a product of two factors from the subgroup \(K\) and one Abelian exponential factor \(A\), typically \(U = K_1 \exp(a) K_2\) with \(a\) in a maximal Abelian subalgebra of the \(-1\)-eigenspace. In quantum circuit synthesis this factorization becomes constructive and recursive, yielding exact decompositions of arbitrary \(n\)-qubit unitaries; in Hamiltonian simulation it yields fixed-depth realizations in which the time dependence is confined to the central Abelian term [2509.05468] [2512.06070]. ## 1. Lie-theoretic formulation For a connected compact semisimple Lie group \(G\) with Lie algebra \(\mathfrak g\), a Cartan involution \(\theta : \mathfrak g \to \mathfrak g\) induces a decomposition \[ \mathfrak g = \mathfrak k \oplus \mathfrak p, \qquad [\mathfrak k,\mathfrak k]\subset \mathfrak k,\; [\mathfrak k,\mathfrak p]\subset \mathfrak p,\; [\mathfrak p,\mathfrak p]\subset \mathfrak k. \] Choosing a maximal Abelian subspace \(\mathfrak a \subset \mathfrak p\), the KAK theorem states that every \(g \in G\) may be written as \[ g = k_1 a k_2, \qquad k_1,k_2 \in K=\exp(\mathfrak k),\; a\in A=\exp(\mathfrak a). \] For simply connected \(G\), the factors are unique up to the action of the Weyl group on \(A\) [2605.10783]. In the quantum-control variant, the decomposition is applied not only to \(SU(2^n)\) as a full group but also to the dynamical Lie algebra of a Hamiltonian. Given \[ H = \sum_k h_k \sigma_k, \qquad \mathfrak g(H)=\mathrm{Lie}\{i\sigma_k\}, \] one chooses \(\mathfrak g = \mathfrak k \oplus \mathfrak m\) with \(iH \in \mathfrak m\), and then a Cartan subalgebra \(h \subset \mathfrak m\). The resulting KHK factorization gives, for time evolution, \[ U(t)=e^{-iHt}=K_c\,e^{-iht}\,K_c^\dagger, \qquad h\in h,\;K_c\in \exp(\mathfrak k), \] so that the circuit depth is independent of \(t\) because the time dependence appears only in the Abelian middle factor [2512.06070]. Two notational conventions coexist in the literature: \((\mathfrak k,\mathfrak p)\) in the symmetric-space formulation and \((\mathfrak l,\mathfrak m)\) in the Khaneja–Glaser recursion. The underlying structure is the same: an involutive splitting of the Lie algebra, a maximal Abelian subalgebra in the \(-1\)-eigenspace, and a group-level \(KAK\) factorization. ## 2. Specialization to \(SU(2^n)\) and recursive Cartan subalgebras A concrete Khaneja–Glaser specialization for \[ \mathfrak g=\mathfrak{su}(2^n) \] chooses \[ \mathfrak l = \mathrm{span}\{\,c\otimes \sigma_z,\; d\otimes \mathbbm 1,\; i z_n \mid c,d\in \mathfrak{su}(2^{n-1})\} \cong \mathfrak{su}(2^{n-1})\oplus \mathfrak{su}(2^{n-1})\oplus u(1), \] and \[ \mathfrak m = \mathrm{span}\{\,a\otimes \sigma_x,\; b\otimes \sigma_y,\; i x_n,\; i y_n \mid a,b\in \mathfrak{su}(2^{n-1})\}, \] with \[ [\mathfrak l,\mathfrak l]\subset \mathfrak l,\qquad [\mathfrak l,\mathfrak m]\subset \mathfrak m,\qquad [\mathfrak m,\mathfrak m]\subset \mathfrak l. \] At the group level, with \(G=SU(2^n)\) and \(K=\exp(\mathfrak l)\cong SU(2^{n-1})\otimes SU(2^{n-1})\otimes U(1)\), every \(U\) admits \[ U = K_1\,\exp(y)\,K_2, \qquad y\in \mathfrak h(n),\;K_i\in K. \] A second Cartan splitting of \((\mathfrak l,\mathfrak l_0)\) refines this to the four-group form \[ U = K_1\,e^{z_1}\,K_2\,e^y\,K_3\,e^{z_2}\,K_4, \quad K_i\in SU(2^{n-1})\otimes U(1),\; y\in \mathfrak h(n),\; z_i\in \mathfrak f(n) \] [2212.12934]. The Cartan subalgebras \(\mathfrak h(n)\) and \(\mathfrak f(n)\) are built recursively. The construction begins with \[ \mathfrak a(2)=i\{x_1x_2,y_1y_2,z_1z_2\},\qquad \mathfrak b(2)=\varnothing, \] then defines \[ \mathfrak s(k)=\bigcup_{i=2}^k\bigl(\mathfrak a(i)\otimes \mathbbm 1^{\,k-i}\bigr), \] followed by \[ \mathfrak a(n)=\{\alpha\otimes \sigma_x,\; ix_n\mid \alpha\in \mathfrak s(n-1)\}, \qquad \mathfrak b(n)=\{\alpha\otimes \sigma_z\mid \alpha\in \mathfrak s(n-1)\}, \] and finally \[ \mathfrak h(n)=\mathrm{span}\{\mathfrak a(n)\}, \qquad \mathfrak f(n)=\mathrm{span}\{\mathfrak b(n)\}. \] This recursion determines which multiqubit generators populate the central Abelian exponentials and which appear in the auxiliary \(z\)-layers [2212.12934]. A complementary formulation describes the same recursive structure through alternating involutions of types AIII, A, and finally AI. In that framework, the first step uses an AIII involution by conjugation with \(Z\) on one qubit, the second uses a block-swap involution on the resulting block-diagonal algebra, and the final two-qubit reduction uses the AI involution in the magic basis. Wierichs et al. show that the Quantum Shannon, Block-ZXZ, and Khaneja–Glaser decompositions implement the same recursive Cartan decomposition [2503.19014]. ## 3. Constructive unitary synthesis The recursive circuit-synthesis algorithm follows the algebraic decomposition directly. At a high level, one first finds a first-level split \[ U = K_1 \exp(y) K_2,\qquad y\in \mathfrak h(n), \] then decomposes the \(K_i\) factors by a second Cartan splitting to obtain the four-group form, and finally recurses on each block in \(SU(2^{n-1})\otimes U(1)\). In practice the recursion is terminated at \(n=2\), where one uses the known optimal three-CNOT factorizations of arbitrary two-qubit gates due to Vatan–Williams and Shende–Markov–Bullock [2212.12934]. The gate-level realization is explicit. Every Cartan generator in \(\mathfrak h(n)\) is a sum of two proportional monomials, \[ \alpha\otimes \sigma_x,\qquad i x_n, \] and the analogous statement for \(\mathfrak f(n)\) uses \(\sigma_z\). These are grouped into two-parameter blocks \[ P_{\pm}(a,b)=\mathrm{diag}(p_\pm,\dots,p_\pm,p_\mp,\dots,p_\mp), \] with \[ p_{\pm}= \begin{pmatrix} \cos(a\pm b) & i\sin(a\pm b)\\ i\sin(a\pm b) & \cos(a\pm b) \end{pmatrix}. \] Each \(P_\pm(a,b)\) is implemented, up to global phase, by two CNOTs framing two one-qubit rotations on the target wire. Appending a tensor \(\otimes \sigma_x\) to the generator corresponds to stamping two extra CNOTs controlled on the \(n\)th qubit. Appending \(\otimes \sigma_z\) corresponds to two fermionic SWAPs between the \(n\)th wire and the appropriate subalgebra wire; each fSWAP can be decomposed into at most four CNOTs or absorbed into relabeling. A lone diagonal generator \(z_1z_2z_n\) is handled by a “Diag” gadget with four CNOTs and one rotation [2212.12934]. This explicit realization is significant because the Cartan factorization is not merely an existence theorem. The algebraic generators themselves are mapped to circuits built from CNOT, SWAP, and one-qubit rotations, and the recursion reaches the level of elementary gates without leaving the Cartan framework. Mansky et al. further emphasize that the construction is independent of the standard CNOT implementation and can be adapted to other cross-qubit circuit elements by changing the block gadget rather than re-deriving the decomposition [2212.12934]. ## 4. Complexity, parameter counts, and comparison with other decompositions For the explicit Khaneja–Glaser construction of \(SU(2^n)\), the CNOT counts for the Abelian factors are \[ \mathcal C_{\mathfrak h}(n) = \#\{\text{CNOTs to build }e^y,\; y\in \mathfrak h(n)\}, \qquad \mathcal C_{\mathfrak f}(n) = \#\{\text{CNOTs to build }e^z,\; z\in \mathfrak f(n)\}. \] Both satisfy the same closed form, \[ \mathcal C_{\mathfrak h}(n) = \mathcal C_{\mathfrak f}(n) = 3\bigl(2^{n-1}+n2^{n-2}\bigr), \] and one full Cartan step uses one \(e^y\) plus two \(e^z\) factors, so \[ \mathcal C_n = \mathcal C_{\mathfrak h}(n)+2\mathcal C_{\mathfrak f}(n) = 3\bigl(2^{n-1}+n2^{n-2}\bigr). \] After adding the recursively decomposed size-\((n-1)\) blocks, \[ T_n=\sum_{i=2}^n 4^{\,n-i}\,\mathcal C_i, \] with \(\mathcal C_2=3\), giving \[ T_n=\frac{21}{16}\,4^n-\mathcal O(n2^n). \] Thus the asymptotic CNOT cost is \(\frac{21}{16}4^n\) [2212.12934]. In the same comparison, earlier synthesis schemes are listed as follows: Barenco et al. give \(\mathcal O(n^3 4^n)\); Knill gives \(\mathcal O(n4^n)\); the Gray-code approach gives \(\mathcal O(4^n)\); the cosine-sine decomposition of Möttönen et al. gives \(4^n-2^{n+1}\); the optimized quantum Shannon decomposition of Shende et al. gives \(\tfrac{23}{48}4^n-\tfrac32 2^n+\tfrac43\); and the theoretical lower bound is \(\left\lceil \tfrac14(4^n-3n-1)\right\rceil\) [2212.12934]. A distinct but related notion of optimality appears in the recursive-Cartan framework of Wierichs et al. There, the alternating AIII\(\to\)A recursion yields a parameter-optimal decomposition: at level \(j\), the chosen Cartan subalgebra has dimension \(2^{n-j-1}\), and summing over levels gives \(2^n-1\) Cartan angles, with the final AI step supplying the remaining three two-qubit parameters. Their analysis unifies several synthesis methods by identifying a common recursive CD rather than by asserting identical gate-count constants [2503.19014]. ## 5. The two-qubit case, Cartan coordinates, and equivalence classes For \(SU(4)\), the Cartan–Khaneja–Glaser decomposition becomes the standard local–nonlocal–local factorization of two-qubit gates. One sets \[ k=\mathrm{Lie}(SU(2)\otimes SU(2)) = \mathrm{span}\{i\sigma_x^1,i\sigma_y^1,i\sigma_z^1,i\sigma_x^2,i\sigma_y^2,i\sigma_z^2\}, \] \[ p=\mathrm{span}\{i\sigma_\alpha^1\sigma_\beta^2:\alpha,\beta\in\{x,y,z\}\}, \] and chooses the maximal Abelian subspace \[ a=\mathrm{span}_{\mathbb R}\{H_1,H_2,H_3\}, \qquad H_1=i\sigma_x^1\sigma_x^2,\; H_2=i\sigma_y^1\sigma_y^2,\; H_3=i\sigma_z^1\sigma_z^2. \] Then every \(U\in SU(4)\) admits \[ U=K_1\,A\,K_2, \qquad A=\exp(\alpha_1 H_1+\alpha_2 H_2+\alpha_3 H_3), \qquad K_1,K_2\in SU(2)\otimes SU(2) \] [2605.10783]. A constructive procedure forms the symmetric matrix \[ M=U^T S U S, \qquad S=\sigma_y\otimes \sigma_y, \] diagonalizes \(M\) by a real orthonormal basis \(O\in SO(4)\), \[ O^T M O = \mathrm{diag}(e^{2i\gamma_1},e^{2i\gamma_2},e^{2i\gamma_3},e^{2i\gamma_4}), \] with \(\gamma_1+\gamma_4=0\) and \(\gamma_2+\gamma_3=0\), and then solves \[ \gamma_1=\alpha_1+\alpha_2+\alpha_3,\quad \gamma_2=\alpha_1-\alpha_2-\alpha_3,\quad \gamma_3=-\alpha_1+\alpha_2-\alpha_3,\quad \gamma_4=-\alpha_1-\alpha_2+\alpha_3. \] This yields the Cartan angles up to Weyl permutations [2605.10783]. A central clarification in recent work concerns equivalence classes. Two distinct notions are separated: double-coset equivalence, \[ U'\!=k_1Uk_2,\qquad k_1,k_2\in SU(2)\otimes SU(2), \] and projective-local equivalence, \[ U'\!=e^{i\phi}k_1Uk_2. \] For double-coset equivalence, the fundamental domain is the tetrahedral cell \[ T=\{(\alpha_1,\alpha_2,\alpha_3)\in \mathbb R^3 \mid \pi/2>\alpha_1\ge \alpha_2\ge |\alpha_3|\ge 0,\; \alpha_1+\alpha_2\le \pi/2\}, \] whereas the usual “Weyl chamber” used in quantum-information practice is recovered only for projective-local equivalence: \[ P=\{(\alpha_1,\alpha_2,\alpha_3) \mid \pi/2>\alpha_1\ge \alpha_2\ge \alpha_3\ge 0,\; \alpha_1+\alpha_2\le \pi/2,\; \text{if }\alpha_3=0\text{ then }\alpha_1\le \pi/4\}. \] This resolves a long-standing inconsistency in the literature on two-qubit local equivalence [2605.10783]. Common gates are placed in these coordinates as \[ I\mapsto (0,0,0),\quad \mathrm{CNOT}\mapsto (\pi/4,0,0),\quad \sqrt{\mathrm{SWAP}}\mapsto (\pi/8,\pi/8,\pi/8), \] \[ \mathrm{iSWAP}\mapsto (\pi/4,\pi/4,0),\quad B\text{-gate}\mapsto (\pi/4,\pi/8,0),\quad \mathrm{QFT}_{2\text{-qubit}}\mapsto (\pi/4,\pi/4,\pi/8) \] [2605.10783]. ## 6. Alternative algebraic constructions and recent extensions One algebraic generalization is the Quotient Algebra Partition framework. In this approach, \(\mathfrak{su}(N)\) is partitioned into Abelian subspaces arranged in conjugate pairs, with commutator closure governed by a binary quotient-algebra rule. For \(2^{p-1}