{
  "id": "moore-penrose-pseudoinverse",
  "title": "Moore–Penrose Pseudoinverse",
  "lang": "en",
  "source_lang": "en",
  "summary": "The Moore–Penrose pseudoinverse extends matrix inversion to rectangular and singular matrices, providing a unique minimum-norm solution to every linear least-squares problem.",
  "category": "mathematics",
  "subcategory": "algebra",
  "type": "concept",
  "aliases": [
    "Moore–Penrose inverse",
    "pseudoinverse",
    "MP inverse"
  ],
  "generated_at": "2026-11-05",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/moore-penrose-pseudoinverse/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/moore-penrose-pseudoinverse/",
      "json": "https://aiwiki.page/zh/moore-penrose-pseudoinverse.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "The Moore–Penrose pseudoinverse of a real or complex matrix \\(A\\) is the unique matrix \\(A^+\\) satisfying four equations that generalize the defining properties of an ordinary matrix inverse. It exists for every finite-dimensional matrix, including rectangular matrices and singular square matrices. Its central role is to produce the minimum-norm solution of a linear least-squares problem, whether the underlying equations have a unique solution, infinitely many solutions, or no exact solution."
    },
    {
      "heading": "Definition",
      "text": "Let \\(A\\in\\mathbb{C}^{m\\times n}\\). Its Moore–Penrose pseudoinverse is the matrix \\(A^+\\in\\mathbb{C}^{n\\times m}\\) satisfying the four Penrose equations:\n\n\\[AA^+A=A,\\]\n\n\\[A^+AA^+=A^+,\\]\n\n\\[(AA^+)^*=AA^+,\\]\n\n\\[(A^+A)^*=A^+A.\\]\n\nHere \\(M^*\\) denotes the conjugate transpose of \\(M\\). For real matrices, it is simply the transpose.\n\nThe first two equations express generalized inverse relationships. The last two require the products \\(AA^+\\) and \\(A^+A\\) to be Hermitian. Together, all four conditions determine \\(A^+\\) uniquely.\n\nThe notation \\(A^\\dagger\\) is also common. The term generalized inverse is broader: a matrix satisfying only \\(AGA=A\\), for example, need not be the Moore–Penrose pseudoinverse and need not be unique.\n\nFor a square nonsingular matrix,\n\n\\[A^+=A^{-1}.\\]\n\nThe pseudoinverse of a zero matrix is the zero matrix of the transposed shape."
    },
    {
      "heading": "Construction by singular value decomposition",
      "text": "The singular value decomposition provides a general construction and establishes existence. Write\n\n\\[A=U\\Sigma V^*,\\]\n\nwhere \\(U\\in\\mathbb{C}^{m\\times m}\\) and \\(V\\in\\mathbb{C}^{n\\times n}\\) are unitary matrices, and \\(\\Sigma\\) is rectangular diagonal. If \\(A\\) has rank \\(r\\), its positive singular values are\n\n\\[\\sigma_1\\geq \\sigma_2\\geq\\cdots\\geq\\sigma_r>0.\\]\n\nForm \\(\\Sigma^+\\) by transposing the shape of \\(\\Sigma\\), replacing each positive singular value by its reciprocal, and leaving zero entries zero. Then\n\n\\[A^+=V\\Sigma^+U^*.\\]\n\nEquivalently, using only the singular vectors associated with positive singular values,\n\n\\[A=\\sum_{i=1}^{r}\\sigma_i u_i v_i^*, \\qquad A^+=\\sum_{i=1}^{r}\\frac{1}{\\sigma_i}v_i u_i^*.\\]\n\nThis formula makes the essential operation explicit: the pseudoinverse reverses every nonzero singular direction and assigns zero to directions that cannot be inverted.\n\nSubstituting this expression into the four Penrose equations verifies existence. For uniqueness, the Hermitian conditions force \\(AA^+\\) and \\(A^+A\\) to be specific orthogonal projections; the remaining conditions then fix how \\(A^+\\) acts on their ranges."
    },
    {
      "heading": "Geometric interpretation",
      "text": "Regard \\(A\\) as a linear map from \\(\\mathbb{C}^n\\) to \\(\\mathbb{C}^m\\), equipped with their standard inner products. The domain decomposes as\n\n\\[\\mathbb{C}^n = \\ker(A)^\\perp\\oplus\\ker(A),\\]\n\nwhere \\(\\ker(A)\\) is the null space and the superscript \\(\\perp\\) denotes the orthogonal complement. The codomain decomposes as\n\n\\[\\mathbb{C}^m = \\operatorname{range}(A)\\oplus\\ker(A^*).\\]\n\nThe restriction\n\n\\[A:\\ker(A)^\\perp\\longrightarrow\\operatorname{range}(A)\\]\n\nis bijective. The pseudoinverse inverts this restricted map and maps \\(\\ker(A^*)\\) to zero.\n\nConsequently,\n\n\\[AA^+=P_{\\operatorname{range}(A)}, \\qquad A^+A=P_{\\operatorname{range}(A^*)},\\]\n\nwhere \\(P_S\\) is the orthogonal projection onto a subspace \\(S\\). The space \\(\\operatorname{range}(A^*)=\\ker(A)^\\perp\\) is the row space, interpreted as a subspace of the domain.\n\nThis interpretation explains why the products are not generally identity matrices: directions outside the relevant ranges cannot be recovered. It also shows that the Moore–Penrose construction depends on the inner products. Changing them can change which solution is regarded as minimum-norm."
    },
    {
      "heading": "Linear equations and least squares",
      "text": "Consider a system of linear equations\n\n\\[Ax=b.\\]\n\nFor every \\(b\\), the vector\n\n\\[x_*=A^+b\\]\n\nminimizes the Euclidean residual:\n\n\\[x_*\\in\\operatorname*{arg\\,min}_{x}\\|Ax-b\\|_2.\\]\n\nAmong all vectors attaining that minimum residual, it is the unique one with the smallest Euclidean norm.\n\nConsistent systems\n\nAn exact solution exists precisely when \\(b\\in\\operatorname{range}(A)\\), equivalently when\n\n\\[AA^+b=b.\\]\n\nIn that case, every exact solution has the form\n\n\\[x=A^+b+(I_n-A^+A)z, \\qquad z\\in\\mathbb{C}^n,\\]\n\nwhere \\(I_n\\) is the identity matrix. The second term ranges over \\(\\ker(A)\\).\n\nSince \\(A^+b\\) is orthogonal to the null space, adding any nonzero null-space component increases the norm. Thus the pseudoinverse selects a single solution even when infinitely many exact solutions exist.\n\nInconsistent systems\n\nIf \\(b\\notin\\operatorname{range}(A)\\), no exact solution exists. Instead,\n\n\\[Ax_*=AA^+b\\]\n\nis the orthogonal projection of \\(b\\) onto the column space of \\(A\\). The residual\n\n\\[b-AA^+b\\]\n\nis orthogonal to that space.\n\nAll least-squares minimizers again have the form\n\n\\[x=A^+b+(I_n-A^+A)z.\\]\n\nThey satisfy the normal equations\n\n\\[A^*Ax=A^*b.\\]\n\nWhen \\(A\\) is rank-deficient, these equations do not uniquely determine \\(x\\); the minimum-norm requirement selects \\(A^+b\\).\n\nExample\n\nFor the single equation\n\n\\[x_1+x_2=1, \\qquad A=\\begin{pmatrix}1&1\\end{pmatrix},\\]\n\nthe pseudoinverse is\n\n\\[A^+= \\begin{pmatrix} 1/2\\\\ 1/2 \\end{pmatrix}.\\]\n\nHence the minimum-norm solution is\n\n\\[x_*= \\begin{pmatrix} 1/2\\\\ 1/2 \\end{pmatrix}.\\]\n\nAll exact solutions are\n\n\\[x= \\begin{pmatrix} 1/2\\\\ 1/2 \\end{pmatrix} +t \\begin{pmatrix} 1\\\\ -1 \\end{pmatrix},\\]\n\nwhose squared norms are \\(1/2+2t^2\\)."
    },
    {
      "heading": "Special formulas and algebraic properties",
      "text": "If \\(A\\) has full column rank, then \\(A^*A\\) is invertible and\n\n\\[A^+=(A^*A)^{-1}A^*.\\]\n\nIf \\(A\\) has full row rank, then \\(AA^*\\) is invertible and\n\n\\[A^+=A^*(AA^*)^{-1}.\\]\n\nNeither formula with an ordinary inverse applies without its rank assumption. General identities using pseudoinverses are\n\n\\[A^+=(A^*A)^+A^* =A^*(AA^*)^+.\\]\n\nOther useful properties include\n\n\\[(A^+)^+=A, \\qquad (A^*)^+=(A^+)^*,\\]\n\n\\[(\\alpha A)^+=\\alpha^{-1}A^+ \\quad\\text{for }\\alpha\\neq0,\\]\n\nand\n\n\\[\\operatorname{rank}(A^+)=\\operatorname{rank}(A).\\]\n\nFor compatible unitary matrices \\(U\\) and \\(V\\),\n\n\\[(UAV)^+=V^*A^+U^*.\\]\n\nThus the construction behaves naturally under orthonormal changes of coordinates. It does not generally obey the reverse-order law\n\n\\[(AB)^+=B^+A^+;\\]\n\nthat identity requires additional conditions. Nor does the pseudoinverse generally distribute over matrix addition.\n\nFor a Hermitian matrix, the spectral theorem gives an analogous construction: retain its eigenvectors, replace each nonzero eigenvalue by its reciprocal, and leave zero eigenvalues zero."
    },
    {
      "heading": "Numerical computation and sensitivity",
      "text": "In numerical linear algebra, the SVD is a standard method for computing pseudoinverses and minimum-norm solutions. Rank-revealing forms of QR decomposition and related factorizations are also used.\n\nIf the aim is only to obtain \\(A^+b\\), explicitly forming the entire matrix \\(A^+\\) is often unnecessary. A factorization can be applied directly to \\(b\\), reducing storage and sometimes computational cost.\n\nThe full-column-rank expression \\((A^*A)^{-1}A^*\\) is useful algebraically but can be numerically undesirable. Forming \\(A^*A\\) squares the spectral condition number:\n\n\\[\\kappa_2(A^*A)=\\kappa_2(A)^2.\\]\n\nThis can magnify the effects of rounding errors.\n\nSmall singular values and numerical rank\n\nIn exact mathematics, every nonzero singular value is inverted. In floating-point arithmetic, extremely small singular values may be indistinguishable from perturbations or rounding error. Software therefore commonly uses a tolerance and treats singular values below it as zero.\n\nThe resulting matrix is the exact pseudoinverse of a truncated approximation to \\(A\\), rather than necessarily the exact pseudoinverse of \\(A\\). The tolerance determines the effective numerical rank.\n\nEven when a small singular value is accurately computed, its reciprocal can amplify measurement noise. Numerical stability of the factorization does not eliminate sensitivity inherent in the problem.\n\nContinuity and rank changes\n\nThe pseudoinverse varies continuously along perturbations that preserve rank, but continuity can fail when rank changes. For example,\n\n\\[A(t)= \\begin{pmatrix} 1&0\\\\ 0&t \\end{pmatrix}.\\]\n\nFor \\(t\\neq0\\),\n\n\\[A(t)^+= \\begin{pmatrix} 1&0\\\\ 0&1/t \\end{pmatrix},\\]\n\nwhereas\n\n\\[A(0)^+= \\begin{pmatrix} 1&0\\\\ 0&0 \\end{pmatrix}.\\]\n\nThe divergence of \\(1/t\\) illustrates why nearly singular problems can be highly sensitive."
    },
    {
      "heading": "Regularization and applications",
      "text": "The pseudoinverse is central to linear regression, constrained linear calculations, and inverse problems. In ordinary least squares, \\(A^+b\\) gives a minimum-norm coefficient vector even when the design matrix is rank-deficient. This algebraic choice does not, by itself, make nonidentifiable coefficients uniquely interpretable.\n\nIn inverse problems, direct pseudoinversion may amplify noise. Regularization replaces exact inversion with a modified problem. For example, minimizing\n\n\\[\\|Ax-b\\|_2^2+\\lambda\\|x\\|_2^2, \\qquad \\lambda>0,\\]\n\ngives\n\n\\[x_\\lambda=(A^*A+\\lambda I_n)^{-1}A^*b.\\]\n\nThis is the basic quadratic penalty used in ridge regression. In singular-vector coordinates, it replaces the reciprocal factor \\(1/\\sigma_i\\) with\n\n\\[\\frac{\\sigma_i}{\\sigma_i^2+\\lambda}.\\]\n\nFor a fixed finite-dimensional matrix,\n\n\\[A^+ = \\lim_{\\lambda\\downarrow0} (A^*A+\\lambda I_n)^{-1}A^*.\\]\n\nTruncating small singular values is another regularization strategy. Both approaches modify the treatment of weakly determined directions; neither should be confused with exact pseudoinversion of the original matrix."
    },
    {
      "heading": "History and scope",
      "text": "The construction is named after Eliakim Hastings Moore, who introduced a generalized inverse in 1920, and Roger Penrose, who gave the four-equation characterization in 1955. It became a standard tool connecting linear algebra, least squares, and numerical computation.\n\nThe finite-dimensional theory extends to operators between Hilbert spaces. For bounded operators with closed range, the Moore–Penrose inverse is a bounded operator with analogous projection properties. Without closed range, a generalized inverse can be unbounded and need not be defined on the entire codomain, so the finite-dimensional existence statement requires qualification in infinite dimensions."
    }
  ],
  "links": [
    "matrix",
    "inverse-matrix",
    "least-squares",
    "conjugate-transpose",
    "singular-value-decomposition",
    "unitary-matrix",
    "matrix-rank",
    "linear-map",
    "null-space",
    "orthogonal-complement",
    "orthogonal-projection",
    "system-of-linear-equations",
    "norm-mathematics",
    "identity-matrix",
    "spectral-theorem",
    "numerical-linear-algebra",
    "qr-decomposition",
    "condition-number",
    "floating-point-arithmetic",
    "linear-regression",
    "inverse-problem",
    "ordinary-least-squares",
    "regularization",
    "ridge-regression",
    "hilbert-space"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}