{
  "id": "adjoint-operator",
  "title": "Adjoint Operator",
  "lang": "en",
  "source_lang": "en",
  "summary": "An adjoint operator transfers a linear operator between the two arguments of an inner product, generalizing the conjugate transpose of a matrix.",
  "category": "mathematics",
  "subcategory": "algebra",
  "type": "concept",
  "aliases": [
    "Hermitian adjoint",
    "Hilbert space adjoint"
  ],
  "generated_at": "2026-11-05",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/adjoint-operator/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/adjoint-operator/",
      "json": "https://aiwiki.page/zh/adjoint-operator.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "An adjoint operator is an operator associated with a linear map that transfers its action from one argument of an inner product to the other. For an operator \\(T:H\\to K\\) between Hilbert spaces, its adjoint \\(T^*:K\\to H\\) is characterized by\n\n\\[\\langle Tx,y\\rangle_K=\\langle x,T^*y\\rangle_H.\\]\n\nIn finite-dimensional spaces with orthonormal coordinates, this is exactly the operation of taking the conjugate transpose of a matrix. In infinite-dimensional spaces, the existence and domain of the adjoint require additional care, especially for unbounded operators. Adjoints are fundamental in functional analysis, differential equations, optimization, and quantum mechanics."
    },
    {
      "heading": "Definition for bounded operators",
      "text": "Let \\(H\\) and \\(K\\) be real or complex Hilbert spaces, and let \\(T:H\\to K\\) be a bounded linear operator. There is a unique bounded linear operator \\(T^*:K\\to H\\) satisfying\n\n\\[\\langle Tx,y\\rangle_K=\\langle x,T^*y\\rangle_H \\qquad (x\\in H,\\ y\\in K).\\]\n\nThroughout this article, complex inner products are taken to be linear in the first argument and conjugate-linear in the second. The alternative convention is also common; the defining adjoint identity remains the same when used consistently.\n\nExistence and uniqueness follow from the Riesz representation theorem: for each fixed \\(y\\in K\\), the map\n\n\\[x\\longmapsto \\langle Tx,y\\rangle_K\\]\n\nis a continuous linear functional on \\(H\\), so it has a unique representation as \\(\\langle x,z\\rangle_H\\) for some \\(z\\in H\\). Setting \\(T^*y=z\\) defines the adjoint.\n\nThe notations \\(T^\\dagger\\) and, less commonly, \\(T^{\\mathrm H}\\) are also used. The dagger notation is particularly common in physics, although in numerical mathematics it can instead denote a pseudoinverse.\n\nThe adjoint depends on the chosen inner products. It is therefore not determined by the underlying linear map alone unless those inner products are specified."
    },
    {
      "heading": "Matrix representation",
      "text": "Suppose \\(H\\) and \\(K\\) are finite-dimensional, and \\(A\\) represents \\(T\\) in orthonormal bases. Then \\(T^*\\) is represented by\n\n\\[A^*=\\overline{A}^{\\,\\mathsf T}.\\]\n\nThus the adjoint is the transpose over the real numbers, and the conjugate transpose over the complex numbers.\n\nFor example,\n\n\\[A= \\begin{pmatrix} 1&i\\\\ 2&3 \\end{pmatrix} \\quad\\Longrightarrow\\quad A^*= \\begin{pmatrix} 1&2\\\\ -i&3 \\end{pmatrix}.\\]\n\nThe orthonormality assumption matters. If coordinate inner products are specified by positive-definite Hermitian Gram matrices \\(G_H\\) and \\(G_K\\), so that\n\n\\[\\langle x,z\\rangle_H=z^*G_Hx, \\qquad \\langle u,y\\rangle_K=y^*G_Ku,\\]\n\nthen the coordinate matrix \\(B\\) of the adjoint is\n\n\\[B=G_H^{-1}A^*G_K.\\]\n\nConsequently, taking a conjugate transpose without accounting for the inner products can give the wrong adjoint in nonorthonormal coordinates.\n\nThe adjoint operator should not be confused with the adjugate matrix, the transpose of the cofactor matrix used in determinant and inverse formulas. Older terminology sometimes calls the adjugate the “classical adjoint.”"
    },
    {
      "heading": "Algebraic and geometric properties",
      "text": "For bounded operators with compatible domains and codomains,\n\n\\[(S+T)^*=S^*+T^*, \\qquad (\\lambda T)^*=\\overline{\\lambda}\\,T^*,\\]\n\n\\[(ST)^*=T^*S^*, \\qquad (T^*)^*=T.\\]\n\nThus taking adjoints reverses the order of composition and is conjugate-linear in the operator. If \\(T\\) has a bounded inverse, then\n\n\\[(T^{-1})^*=(T^*)^{-1}.\\]\n\nFor the operator norm,\n\n\\[\\|T^*\\|=\\|T\\|, \\qquad \\|T^*T\\|=\\|T\\|^2.\\]\n\nMoreover,\n\n\\[\\langle x,T^*Tx\\rangle=\\|Tx\\|^2\\geq 0.\\]\n\nThe operator \\(T^*T\\) is therefore self-adjoint and positive.\n\nAdjoints also express the relationship between an operator’s kernel and range. Writing \\(\\operatorname{ran}T\\) for the range,\n\n\\[\\ker T^*=(\\operatorname{ran}T)^\\perp, \\qquad \\ker T=(\\operatorname{ran}T^*)^\\perp,\\]\n\nwhere \\({}^\\perp\\) denotes the orthogonal complement. Equivalently,\n\n\\[\\overline{\\operatorname{ran}T}=(\\ker T^*)^\\perp, \\qquad \\overline{\\operatorname{ran}T^*}=(\\ker T)^\\perp.\\]\n\nThe closures cannot generally be omitted in infinite dimensions. For example, \\(T\\) has dense range exactly when \\(T^*\\) is injective."
    },
    {
      "heading": "Important classes of operators",
      "text": "Several major classes of operators are defined through their adjoints.\n\n- A self-adjoint operator satisfies \\(T=T^*\\).\n- A skew-adjoint operator satisfies \\(T^*=-T\\).\n- A normal operator satisfies \\(T^*T=TT^*\\).\n- A unitary operator \\(U:H\\to K\\) satisfies \\(U^*U=I_H\\) and \\(UU^*=I_K\\).\n- An isometry satisfies \\(T^*T=I_H\\), but need not be onto.\n- An orthogonal projection satisfies \\(P^2=P=P^*\\).\n\nFor real Hilbert spaces, surjective inner-product-preserving operators are usually called orthogonal rather than unitary.\n\nThese identities connect adjoints to the spectral theorem. In finite-dimensional complex inner-product spaces, normal operators admit an orthonormal basis of eigenvectors. Self-adjoint operators have real eigenvalues, while unitary operators have eigenvalues of absolute value one.\n\nFor a general matrix \\(A\\), the singular value decomposition is governed by the positive operators \\(A^*A\\) and \\(AA^*\\). Their nonzero eigenvalues agree, and their square roots are the nonzero singular values of \\(A\\)."
    },
    {
      "heading": "Examples",
      "text": "Multiplication operators\n\nOn the space \\(L^2(X,\\mu)\\), with\n\n\\[\\langle f,g\\rangle=\\int_X f\\,\\overline g\\,d\\mu,\\]\n\nlet \\(M_mf=mf\\), where \\(m\\) is an essentially bounded measurable function. Then\n\n\\[M_m^*=M_{\\overline m}.\\]\n\nIn particular, \\(M_m\\) is self-adjoint exactly when \\(m\\) is real-valued almost everywhere.\n\nIntegral operators\n\nSuppose \\(X\\) and \\(Y\\) are finite measure spaces and \\(k\\in L^2(Y\\times X)\\). The integral operator\n\n\\[(Tf)(y)=\\int_X k(y,x)f(x)\\,d\\mu(x)\\]\n\ndefines a bounded map from \\(L^2(X,\\mu)\\) to \\(L^2(Y,\\nu)\\). Its adjoint is\n\n\\[(T^*g)(x)=\\int_Y\\overline{k(y,x)}\\,g(y)\\,d\\nu(y).\\]\n\nThus taking an adjoint exchanges the input and output variables and conjugates the kernel, paralleling the conjugate transpose of a matrix.\n\nThe unilateral shift\n\nOn the Hilbert space \\(\\ell^2(\\mathbb N_0)\\), define\n\n\\[S(x_0,x_1,x_2,\\ldots)=(0,x_0,x_1,\\ldots).\\]\n\nThen\n\n\\[S^*(x_0,x_1,x_2,\\ldots)=(x_1,x_2,x_3,\\ldots).\\]\n\nHere \\(S^*S=I\\), but \\(SS^*\\neq I\\): the latter removes the zeroth coordinate. This illustrates why an isometry need not be unitary in infinite dimensions."
    },
    {
      "heading": "Unbounded operators and their domains",
      "text": "Many operators arising from differential equations are unbounded. Such an operator is specified not merely by a rule but also by a linear domain:\n\n\\[T:D(T)\\subseteq H\\longrightarrow K.\\]\n\nAssume \\(D(T)\\) is dense in \\(H\\). The domain of the adjoint is\n\n\\[D(T^*)= \\left\\{ y\\in K: \\text{there exists }z\\in H \\text{ such that } \\langle Tx,y\\rangle_K=\\langle x,z\\rangle_H \\text{ for all }x\\in D(T) \\right\\}.\\]\n\nDensity makes \\(z\\) unique, and \\(T^*y\\) is defined to be this vector.\n\nEquivalently, \\(y\\in D(T^*)\\) precisely when\n\n\\[x\\longmapsto\\langle Tx,y\\rangle_K\\]\n\nis bounded on \\(D(T)\\) with respect to the norm of \\(H\\).\n\nUnlike the bounded case, \\(D(T^*)\\) need not be all of \\(K\\), or even dense. The adjoint is always a closed operator. A densely defined operator \\(T\\) is closable exactly when \\(D(T^*)\\) is dense; in that case,\n\n\\[T^{**}=\\overline T,\\]\n\nwhere \\(\\overline T\\) denotes its closure.\n\nFor unbounded operators, equality includes equality of domains. Merely showing\n\n\\[\\langle Tx,y\\rangle=\\langle x,Ty\\rangle \\qquad(x,y\\in D(T))\\]\n\nestablishes that \\(T\\) is symmetric, meaning \\(T\\subseteq T^*\\). Self-adjointness requires the stronger statement\n\n\\[D(T)=D(T^*), \\qquad Tx=T^*x\\quad(x\\in D(T)).\\]\n\nThis distinction is essential in infinite-dimensional spectral theory.\n\nDifferentiation and boundary conditions\n\nConsider\n\n\\[Tf=f'\\]\n\non \\(L^2(0,1)\\), with\n\n\\[D(T)=H_0^1(0,1),\\]\n\nthe Sobolev space of \\(H^1\\) functions whose endpoint traces vanish. By integration by parts,\n\n\\[\\int_0^1 f'(x)\\overline{g(x)}\\,dx = [f(x)\\overline{g(x)}]_0^1 - \\int_0^1 f(x)\\overline{g'(x)}\\,dx.\\]\n\nThe boundary term vanishes for every \\(f\\in D(T)\\), yielding\n\n\\[T^*g=-g', \\qquad D(T^*)=H^1(0,1).\\]\n\nThe adjoint therefore has a different domain from the original operator. If instead \\(f'\\) is defined on all of \\(H^1(0,1)\\), its adjoint is \\(-g'\\) on \\(H_0^1(0,1)\\).\n\nA formal adjoint is the differential expression obtained by integration by parts before domains and boundary conditions are fully specified. It need not be the complete Hilbert-space adjoint: the latter includes its precise domain."
    },
    {
      "heading": "Relation to dual-space adjoints",
      "text": "There is a related construction that does not require an inner product. For a bounded linear map \\(T:X\\to Y\\) between Banach spaces, its transpose, also called its Banach-space adjoint, is\n\n\\[T':Y'\\to X', \\qquad T'\\varphi=\\varphi\\circ T,\\]\n\nwhere \\(X'\\) and \\(Y'\\) are the continuous dual spaces.\n\nFor complex spaces, this transpose is linear in \\(T\\):\n\n\\[(\\lambda T)'=\\lambda T',\\]\n\nwhereas the Hilbert-space adjoint satisfies\n\n\\[(\\lambda T)^*=\\overline{\\lambda}\\,T^*.\\]\n\nThe two constructions are connected by the Riesz identifications. Under the inner-product convention used here, define the conjugate-linear maps\n\n\\[J_Hx=\\langle\\,\\cdot\\,,x\\rangle_H, \\qquad J_Ky=\\langle\\,\\cdot\\,,y\\rangle_K.\\]\n\nThen\n\n\\[T^*=J_H^{-1}T'J_K.\\]\n\nThe conjugate-linearity of these identifications explains why the Hilbert-space adjoint introduces complex conjugation."
    },
    {
      "heading": "Applications",
      "text": "Least squares and optimization\n\nFor a bounded operator \\(T:H\\to K\\) and data \\(b\\in K\\), consider the least-squares objective\n\n\\[F(x)=\\frac12\\|Tx-b\\|^2.\\]\n\nIts gradient is\n\n\\[\\nabla F(x)=T^*(Tx-b).\\]\n\nIn a complex Hilbert space, this statement uses the underlying real inner product \\(\\operatorname{Re}\\langle\\cdot,\\cdot\\rangle\\).\n\nAny minimizer satisfies the normal equation\n\n\\[T^*Tx=T^*b.\\]\n\nConversely, every solution of this equation minimizes \\(F\\). In infinite dimensions, a minimizer need not exist if the range of \\(T\\) is not closed.\n\nAdjoint-state methods extend the same principle to optimization constrained by partial differential equations. They transfer derivatives through the governing equations, allowing gradients of scalar objectives to be computed without separately differentiating the full state with respect to every parameter.\n\nQuantum mechanics\n\nIn quantum mechanics, observables are represented by self-adjoint operators, often unbounded ones. Domain conditions are part of their mathematical specification: a formally symmetric differential expression alone does not determine a self-adjoint observable. Unitary evolution also uses the adjoint through \\(U^*=U^{-1}\\).\n\nNumerical computation\n\nAlgorithms for large-scale inverse problems often require only the actions \\(x\\mapsto Tx\\) and \\(y\\mapsto T^*y\\), rather than an explicitly stored matrix. An implementation can be checked using the adjoint identity\n\n\\[\\langle Tx,y\\rangle_K=\\langle x,T^*y\\rangle_H\\]\n\non test vectors, allowing for floating-point error. If discretized spaces use weighted inner products, those weights must be included; an ordinary matrix transpose may not represent the intended adjoint."
    }
  ],
  "links": [
    "linear-map",
    "inner-product",
    "hilbert-space",
    "conjugate-transpose",
    "functional-analysis",
    "linear-functional",
    "orthonormal-basis",
    "matrix-transpose",
    "gram-matrix",
    "norm-mathematics",
    "orthogonal-complement",
    "self-adjoint-operator",
    "orthogonal-projection",
    "spectral-theorem",
    "singular-value-decomposition",
    "integral",
    "differential-equation",
    "sobolev-space",
    "integration-by-parts",
    "banach-space",
    "dual-space",
    "least-squares",
    "gradient",
    "partial-differential-equation",
    "quantum-mechanics",
    "inverse-problem"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}