{
  "id": "pushforward-measure",
  "title": "Pushforward Measure",
  "lang": "en",
  "source_lang": "en",
  "summary": "A pushforward measure transfers a measure through a measurable map by assigning each target set the measure of its preimage.",
  "category": "mathematics",
  "subcategory": "analysis",
  "type": "concept",
  "aliases": [
    "Image measure",
    "Induced measure",
    "Direct image measure"
  ],
  "generated_at": "2026-11-05",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/pushforward-measure/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/pushforward-measure/",
      "json": "https://aiwiki.page/zh/pushforward-measure.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "A pushforward measure is the measure obtained by transporting an existing measure along a measurable map. If \\(T:X\\to Y\\) maps one measurable space into another and \\(\\mu\\) is a measure on \\(X\\), its pushforward assigns a measurable set \\(B\\subseteq Y\\) the value \\(\\mu(T^{-1}(B))\\). Usually written \\(T_{\\#}\\mu\\) or \\(T_*\\mu\\), this construction is fundamental in measure theory and probability: the distribution of a random variable is the pushforward of the underlying probability measure."
    },
    {
      "heading": "Definition",
      "text": "Let \\((X,\\mathcal A)\\) and \\((Y,\\mathcal B)\\) be measurable spaces, and let \\(\\mu\\) be a measure on \\(\\mathcal A\\). A map \\(T:X\\to Y\\) is measurable when\n\n\\[T^{-1}(B)\\in\\mathcal A \\qquad\\text{for every }B\\in\\mathcal B.\\]\n\nThe pushforward of \\(\\mu\\) under \\(T\\) is the measure on \\((Y,\\mathcal B)\\) defined by\n\n\\[(T_{\\#}\\mu)(B)=\\mu\\bigl(T^{-1}(B)\\bigr), \\qquad B\\in\\mathcal B.\\]\n\nHere\n\n\\[T^{-1}(B)=\\{x\\in X:T(x)\\in B\\}\\]\n\nis a preimage, not an inverse function. Consequently, \\(T\\) need not be injective, surjective, continuous, or differentiable.\n\nThe sigma-algebra \\(\\mathcal B\\) is part of the construction. The pushforward is defined on its sets, not automatically on every subset of \\(Y\\).\n\nWhy this defines a measure\n\nPreimages preserve the empty set and countable unions. They also preserve disjointness: if \\(B_1,B_2,\\ldots\\) are pairwise disjoint, then their preimages are pairwise disjoint. Therefore,\n\n\\[\\begin{aligned} (T_{\\#}\\mu)\\left(\\bigcup_{n=1}^{\\infty}B_n\\right) &=\\mu\\left(\\bigcup_{n=1}^{\\infty}T^{-1}(B_n)\\right)\\\\ &=\\sum_{n=1}^{\\infty}\\mu(T^{-1}(B_n))\\\\ &=\\sum_{n=1}^{\\infty}(T_{\\#}\\mu)(B_n). \\end{aligned}\\]\n\nTogether with \\((T_{\\#}\\mu)(\\varnothing)=0\\), this establishes the measure axioms.\n\nThe definition uses preimages because they have these properties regardless of whether \\(T\\) is invertible. Simply assigning mass using images of sets would not provide an analogous general construction: images of disjoint sets can overlap, and images of measurable sets need not be measurable."
    },
    {
      "heading": "Basic properties",
      "text": "Preservation of total mass\n\nSince \\(T^{-1}(Y)=X\\),\n\n\\[(T_{\\#}\\mu)(Y)=\\mu(X).\\]\n\nThus finite measures remain finite, and probability measures remain probability measures.\n\nMore generally, the construction is linear for nonnegative combinations of measures:\n\n\\[T_{\\#}(a\\mu+b\\nu) =aT_{\\#}\\mu+bT_{\\#}\\nu, \\qquad a,b\\geq0.\\]\n\nComposition\n\nIf \\(S:Y\\to Z\\) is another measurable map, then\n\n\\[(S\\circ T)_{\\#}\\mu=S_{\\#}(T_{\\#}\\mu).\\]\n\nThis follows from\n\n\\[(S\\circ T)^{-1}(C)=T^{-1}(S^{-1}(C)).\\]\n\nThe identity map also satisfies\n\n\\[(\\operatorname{id}_X)_{\\#}\\mu=\\mu.\\]\n\nThese identities make pushforward compatible with function composition: transporting a measure through two maps in succession gives the same result as transporting it through their composition.\n\nChanges on null sets\n\nIf two measurable maps \\(T,S:X\\to Y\\) agree outside a measurable set of \\(\\mu\\)-measure zero, then\n\n\\[T_{\\#}\\mu=S_{\\#}\\mu.\\]\n\nThus modifying a map almost everywhere does not alter its pushforward."
    },
    {
      "heading": "Integration against a pushforward",
      "text": "The central integration identity is\n\n\\[\\boxed{ \\int_Y f(y)\\,d(T_{\\#}\\mu)(y) = \\int_X f(T(x))\\,d\\mu(x) }\\]\n\nfor every nonnegative measurable function \\(f:Y\\to[0,\\infty]\\). It also holds for real- or complex-valued integrable functions.\n\nIn particular,\n\n\\[\\int_Y |f|\\,d(T_{\\#}\\mu) = \\int_X |f\\circ T|\\,d\\mu,\\]\n\nso \\(f\\) is integrable with respect to \\(T_{\\#}\\mu\\) exactly when \\(f\\circ T\\) is integrable with respect to \\(\\mu\\).\n\nThe identity is first checked for indicator functions:\n\n\\[\\int_Y \\mathbf 1_B\\,d(T_{\\#}\\mu) =(T_{\\#}\\mu)(B) =\\int_X\\mathbf 1_B(T(x))\\,d\\mu(x).\\]\n\nIt then extends to nonnegative simple functions and, by monotone approximation, to nonnegative measurable functions. Integrable functions follow by decomposition.\n\nThis is an abstract change-of-variables formula for the Lebesgue integral. It requires no derivatives, Jacobians, or invertibility. Those enter only when expressing the pushforward relative to a particular reference measure.\n\nThe identity also shows a useful duality: measures move forward under \\(T\\), while functions are composed backward with \\(T\\)."
    },
    {
      "heading": "Probability distributions",
      "text": "Let \\((\\Omega,\\mathcal F,\\mathbb P)\\) be a probability space, and let \\(X:\\Omega\\to S\\) be a random variable or random element. Its probability distribution, also called its law, is\n\n\\[\\mathcal L(X)=X_{\\#}\\mathbb P.\\]\n\nFor each measurable \\(B\\subseteq S\\),\n\n\\[\\mathcal L(X)(B) =\\mathbb P(X\\in B).\\]\n\nIf \\(g:S\\to E\\) is measurable, composition gives\n\n\\[\\mathcal L(g(X))=g_{\\#}\\mathcal L(X).\\]\n\nThus the distribution of a transformed random variable depends only on its original distribution and the transformation, not on the particular probability space used to realize it.\n\nThe integration identity becomes\n\n\\[\\mathbb E[g(X)] =\\int_S g(x)\\,d\\mathcal L(X)(x),\\]\n\nwhenever the expectation is defined.\n\nFor real-valued \\(X\\), its cumulative distribution function is\n\n\\[F_X(t) =(X_{\\#}\\mathbb P)((-\\infty,t]).\\]\n\nA joint distribution is likewise a pushforward under the map\n\n\\[\\omega\\longmapsto(X(\\omega),Y(\\omega)).\\]\n\nIts marginal distributions are obtained by pushing that joint measure forward under the coordinate projections."
    },
    {
      "heading": "Examples",
      "text": "Discrete measures\n\nIf a finite measure is concentrated on points \\(x_1,\\ldots,x_n\\), with masses \\(p_1,\\ldots,p_n\\), then\n\n\\[(T_{\\#}\\mu)(B) =\\sum_{\\{i:T(x_i)\\in B\\}}p_i.\\]\n\nPoints with the same image contribute to the same target mass.\n\nFor example, suppose \\(\\mu\\) assigns masses \\(0.2,0.3,0.5\\) to \\(-1,0,1\\), respectively. Under \\(T(x)=x^2\\), the pushforward assigns mass \\(0.3\\) to \\(0\\) and mass \\(0.7\\) to \\(1\\).\n\nA constant map\n\nIf \\(T(x)=y_0\\) and \\(\\mu(X)=m<\\infty\\), then\n\n\\[(T_{\\#}\\mu)(B) = \\begin{cases} m,&y_0\\in B,\\\\ 0,&y_0\\notin B. \\end{cases}\\]\n\nIn particular, any probability measure can be pushed forward to a measure concentrated at a single point. Pushforward therefore need not preserve the absence of atoms.\n\nSquaring a uniform random variable\n\nLet \\(U\\) have the uniform distribution on \\([-1,1]\\), and set \\(Y=U^2\\). For \\(0\\leq y\\leq1\\),\n\n\\[\\mathbb P(Y\\leq y) =\\mathbb P(-\\sqrt y\\leq U\\leq\\sqrt y) =\\sqrt y.\\]\n\nConsequently, \\(Y\\) has density\n\n\\[f_Y(y)=\\frac{1}{2\\sqrt y}, \\qquad 0<y<1,\\]\n\nand zero density outside that interval. The singularity at zero is integrable and does not represent an atom."
    },
    {
      "heading": "Densities and Jacobian formulas",
      "text": "A pushforward measure always exists under the measurability assumptions, but it need not have a density. Density formulas require additional hypotheses.\n\nSuppose \\(U,V\\subseteq\\mathbb R^d\\) are open, \\(T:U\\to V\\) is a continuously differentiable bijection with continuously differentiable inverse, and\n\n\\[d\\mu(x)=\\rho(x)\\,dx,\\]\n\nwhere \\(dx\\) denotes Lebesgue measure. Then\n\n\\[d(T_{\\#}\\mu)(y) = \\rho(T^{-1}(y)) \\left|\\det DT^{-1}(y)\\right|\\,dy.\\]\n\nEquivalently,\n\n\\[\\rho_{T_{\\#}\\mu}(y) = \\frac{\\rho(T^{-1}(y))} {\\left|\\det DT(T^{-1}(y))\\right|}.\\]\n\nHere \\(DT\\) is the Jacobian matrix, and its determinant describes local volume scaling.\n\nFor an invertible affine transformation \\(T(x)=Ax+b\\), this becomes\n\n\\[\\rho_{T_{\\#}\\mu}(y) = \\frac{\\rho(A^{-1}(y-b))}{|\\det A|}.\\]\n\nIn one dimension, a non-injective map can often be handled by summing over inverse branches:\n\n\\[\\rho_{T_{\\#}\\mu}(y) = \\sum_{x:T(x)=y}\\frac{\\rho(x)}{|T'(x)|}.\\]\n\nFor example, this formula holds almost everywhere when the source density is supported on finitely many open intervals on each of which \\(T\\) is continuously differentiable with nonzero derivative. Each preimage branch contributes separately.\n\nIf a map collapses a positive amount of mass onto a point or a lower-dimensional set, the pushforward can have a singular component instead of an ordinary density."
    },
    {
      "heading": "Limitations and measure-theoretic subtleties",
      "text": "Sigma-finiteness need not be preserved\n\nA measure is sigma-finite if its space is a countable union of sets of finite measure. Even when \\(\\mu\\) is sigma-finite, \\(T_{\\#}\\mu\\) need not be.\n\nFor example, let \\(\\mu\\) be Lebesgue measure on \\(\\mathbb R\\), and map every real number to a single point. The resulting one-point space has infinite measure, and its only finite-measure measurable set is the empty set. Its measure is therefore not sigma-finite.\n\nAbsolute continuity can be lost\n\nIf \\(\\mu\\) has a density relative to Lebesgue measure, a constant map still sends it to a point mass. Thus the absolute continuity needed for a density is not automatically preserved.\n\nThere is, however, a relative preservation property: if \\(\\mu\\ll\\nu\\), then\n\n\\[T_{\\#}\\mu\\ll T_{\\#}\\nu.\\]\n\nIndeed, \\(T_{\\#}\\nu(B)=0\\) implies \\(\\nu(T^{-1}(B))=0\\), hence \\(\\mu(T^{-1}(B))=0\\).\n\nInformation is generally lost\n\nA non-injective map merges points, so its pushforward generally cannot determine the original measure. For \\(T(x)=x^2\\), point masses at \\(1\\) and \\(-1\\) have the same pushforward.\n\nBy contrast, if \\(T\\) is a bijection and both \\(T\\) and \\(T^{-1}\\) are measurable, then\n\n\\[(T^{-1})_{\\#}(T_{\\#}\\mu)=\\mu.\\]"
    },
    {
      "heading": "Applications",
      "text": "Measure-preserving dynamics\n\nA measurable transformation \\(T:X\\to X\\) preserves a measure \\(\\mu\\) precisely when\n\n\\[T_{\\#}\\mu=\\mu,\\]\n\nor equivalently,\n\n\\[\\mu(T^{-1}(A))=\\mu(A)\\]\n\nfor every measurable \\(A\\). This is the invariant-measure condition used in the study of a dynamical system.\n\nOptimal transport\n\nIn optimal transport, a map \\(T\\) transports a source probability measure \\(\\mu\\) to a target probability measure \\(\\nu\\) when\n\n\\[T_{\\#}\\mu=\\nu.\\]\n\nThe map may be chosen to minimize a cost such as\n\n\\[\\int_X c(x,T(x))\\,d\\mu(x).\\]\n\nA deterministic transport map need not exist: a source concentrated at one point cannot be mapped deterministically to a target distributing mass over several points. Transport plans extend the formulation by allowing mass from one source point to be split among multiple destinations.\n\nWeak convergence\n\nSuppose probability measures \\(\\mu_n\\) converge weakly to \\(\\mu\\) on a metric space, and \\(T\\) is continuous into another metric space. Then\n\n\\[T_{\\#}\\mu_n\\Rightarrow T_{\\#}\\mu.\\]\n\nFor every bounded continuous function \\(f\\) on the target, \\(f\\circ T\\) is bounded and continuous on the source, so\n\n\\[\\int f\\,d(T_{\\#}\\mu_n) = \\int f\\circ T\\,d\\mu_n \\longrightarrow \\int f\\circ T\\,d\\mu = \\int f\\,d(T_{\\#}\\mu).\\]\n\nThis is the measure-theoretic form of the continuous mapping principle used in probability limit theorems."
    }
  ],
  "links": [
    "measurable-function",
    "measure-theory",
    "measurable-space",
    "inverse-function",
    "sigma-algebra",
    "function-composition",
    "almost-everywhere",
    "indicator-function",
    "lebesgue-integral",
    "probability-space",
    "random-variable",
    "probability-distribution",
    "expected-value",
    "cumulative-distribution-function",
    "joint-probability-distribution",
    "uniform-distribution",
    "probability-density-function",
    "lebesgue-measure",
    "jacobian-matrix",
    "determinant",
    "absolute-continuity",
    "dynamical-system",
    "continuous-function"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}