On Linear Transformations of Intersections

Abstract

For any linear transformation and two convex closed sets, we provide necessary and sufficient conditions for the transformation of the intersection of the sets to coincide with the intersection of their images. We also identify conditions for non-convex closed sets, continuous transformations, and multiple sets. We demonstrate the usefulness of our results via an application to the economics literature of mechanism design.

This is a preview of subscription content, access via your institution.

References

  1. 1.

    Aumann, R.J., Hart, S.: Bi-convexity and bi-martingales. Israel J. Math. 54 (2), 159–180 (1986)

    MathSciNet  Article  Google Scholar 

  2. 2.

    Axler, S.J.: Linear Algebra Done Right. Springer, New York (1997)

    Book  Google Scholar 

  3. 3.

    Breen, M.: Starshaped unions and nonempty intersections of convex sets in \(\mathbb {R}^{d}\). Proc. Amer. Math. Soc. 108(3), 817–820 (1990)

    MathSciNet  MATH  Google Scholar 

  4. 4.

    Bressan, A.: Directional convexity and finite optimality conditions. J. Math. Anal. Appl. 125, 234–246 (1987)

    MathSciNet  Article  Google Scholar 

  5. 5.

    Gershkov, A., Goeree, J.K., Kushnir, A., Moldovanu, B., Shi, X.: On the equivalence of Bayesian and dominant strategy implementation. Econometrica 81, 197–220 (2013)

    MathSciNet  Article  Google Scholar 

  6. 6.

    Goeree, J.K., Kushnir, A.: A Geometric Approach to Mechanism Design Working Paper, University of New South Wales and Tepper School of Business, Carnegie Mellon University (2017)

  7. 7.

    Hiriart-Urruty, J.B., Lemaréchal, C.: Fundamentals of Convex Analysis. Springer Science & Business Media (2012)

  8. 8.

    Holtzman, J.M., Halkin, H.: Directional convexity and the maximum principle for discrete systems. SIAM J. Control 4(2), 263–275 (1966)

    MathSciNet  Article  Google Scholar 

  9. 9.

    Horvath, C., Lassonde, M.: Intersection of sets with -connected unions. Proc. Amer. Math. Soc. 125(4), 1209–1214 (1997)

    MathSciNet  Article  Google Scholar 

  10. 10.

    Kushnir, A.: On sufficiency of dominant strategy implementation in environments with correlated types. Econ. Lett. 133, 4–6 (2015)

    MathSciNet  Article  Google Scholar 

  11. 11.

    Kushnir, A., Liu, S.: On the equivalence of Bayesian and dominant strategy implementation for environments with nonlinear utilities. Econ. Theory 67, 617–644 (2019)

    MathSciNet  Article  Google Scholar 

  12. 12.

    Laffont, J.-J., Maskin, E.: A differential approach to dominant strategy mechanisms. Econometrica 48, 1507–1520 (1980)

    MathSciNet  Article  Google Scholar 

  13. 13.

    Manelli, A.M., Vincent, D.R.: Bayesian and dominant-strategy implementation in the independent private values model. Econometrica 78, 1905–1938 (2010)

    MathSciNet  Article  Google Scholar 

  14. 14.

    Milgrom, P.R.: Putting Auction Theory to Work. Cambridge University Press, Cambridge (2004)

  15. 15.

    Munkres, J.R.: Topology, 2nd edn. Prentice Hall Inc., USA (2000)

    MATH  Google Scholar 

  16. 16.

    Myerson, R.: Incentive compatibility and the bargaining problem. Econometrica 47, 61–73 (1979)

    MathSciNet  Article  Google Scholar 

  17. 17.

    Rockafellar, R.T.: Convex Analysis. Princeton University Press, Princeton (1997)

  18. 18.

    Roth, A.E.: What have we learned from market design?. Econ. J. 118(527), 285–310 (2008)

    Article  Google Scholar 

Download references

Acknowledgments

We are very grateful to Itai Ashlagi, Egon Balas, Heinz Bauschke, Jérôme Bolte, Boris Bukh, Keenan Crane, Patrick Combettes, Gerard Cornuejols, Federico Echenique, Alfred Galichon, Ben Golub, Sergiu Hart, Fatma Kılınç-Karzan, Michael McCoy, Javier Pena, Marek Pycia, R Ravi, Stephen Spear, Rakesh Vohra, Josephine Yu, Weijie Zhong, two anonymous referees, and seminar participants at Columbia University, Carnegie Mellon University, Higher School of Economics, New Economic School, and the University of Pittsburgh for discussions and useful suggestions. Shuo Liu acknowledges the financial support by the Forschungskredit of the University of Zurich, grant no. FK-17-018.

Author information

Affiliations

Authors

Corresponding author

Correspondence to Alexey Kushnir.

Additional information

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Appendix

Appendix

Theorem 1

Consider two convex closed sets\(A, B\subset \mathbb {R}^{n}\). If for all linear transformations\(T:\mathbb {R}^{n}\rightarrow \mathbb {R}^{n-1}\)we have T(AB) = TATB, then AB is convex.

Proof

Take some aA and bB such that ab, and consider any linear transformation \(T: \mathbb {R}^{n}\rightarrow \mathbb {R}^{n-1}\) with \(\ker (T)=\{v\in \mathbb {R}^{n}: v=k\cdot (a-b), k\in \mathbb {R}\}\). By construction, the kernel of T amounts to the lines spanned by ab, and we have \(a-b\in \ker (T)\) and Ta = Tb = t for some \(t\in \mathbb {R}^{n-1}\). As tTATB = T(AB), there exists cAB such that Tc = t. As \(\dim (\ker (T))=1\), points a, b, c must lie on one straight line. As A and B are convex and cAB, we have [a, c] ⊂ A and [b, c] ⊂ B, which implies that [a, b] ⊂ AB. Hence, AB is convex. □

Proof of Theorem 8

We want to prove that if \(\forall d\in \ker (T)^{\perp }~E(A\cap B, d)\subseteq E(A, d)\cup E(B, d)\), then T(AB) = TATB. As T(AB) ⊂ TATB and both T(AB) and TATB are convex compact sets, it suffices to show that E(T(AB), d) ⊂ E(TATB, d) for all directions \(d\in \mathbb {R}^{m}\), that is, every support point of T(AB) is also a support point of TATB in the same direction. For this purpose, consider any direction \(d\in \mathbb {R}^{m}\). Let \(T^{*}: \mathbb {R}^{m}\rightarrow \mathbb {R}^{n}\) be the adjoint operator (or the transpose) of T. As \(\text {image}(T^{*})=\ker (T)^{\perp }\) (see p. 120 in [2]) we have \(T^{*}(d)\in \ker (T)^{\perp }\). The condition of the theorem then ensures that \(E(A\cap B, T^{*}(d))\subseteq E(A, T^{*}(d))\cup E(B, T^{*}(d))\). By the definition of adjoint operators, we further have \( E(T(A\cap B), d)\subseteq E(TA, d)\cup E(TB, d). \)

Now consider any \(t\in E(T(A\cap B), d)\subseteq E(TA, d)\cup E(TB, d)\). Without loss, we suppose that tE(TA, d). T(AB) ⊂ TATB implies that tTATB, and since TATB is a subset of TA, from the relation tE(TA, d) we can conclude that tE(TATB, d). □

Proof of Theorem 9

We want to prove that T(AB) = TATB if and only if \(\forall d\in \ker (T)^{\perp }\) and ∀uE(AB, d), \(\exists d^{\prime }, d^{\prime \prime }\in \ker (T)^{\perp }\) such that \(d^{\prime }+d^{\prime \prime }=d\) and \(u\in E(A, d^{\prime })\cap E(B, d^{\prime \prime })\).

(If statement) For arbitrary direction \(z\in \mathbb {R}^{m}\) the support function of the image of the intersection equals

$$ S^{T(A\cap B)}(z)=\sup_{t\in T(A\cap B)} t\cdot z=\sup_{x\in A\cap B} x\cdot T^{*}(z)=S^{A\cap B}(T^{*}(z)), $$

where T is the adjoint operator. As \(\text {image}(T^{*})=\ker (T)^{\perp }\) we have \(T^{*}(z)\in \ker (T)^{\perp }\). The support function for the intersection of convex sets having non-empty intersection of their relative interiors (riA ∩riB) can be conveniently characterized by (see Corollary 16.4.1 in [17])

$$ S^{A\cap B}(T^{*}(z))=\underset{d^{\prime},d^{\prime\prime} \in \mathbb{R}^{n} }{\underset{d^{\prime}+d^{\prime\prime}=T^{*}(z)}{\inf}} (S^{A}(d^{\prime})+S^{B}(d^{\prime\prime})). $$
(1)

Since set AB is compact, the set of support points E(AB, T(z)) is non-empty. The condition of the theorem then implies that for any uE(AB, T(z)), there exist \(d^{\prime }, d^{\prime \prime }\in \ker (T)^{\perp }\), such that \(d^{\prime }+d^{\prime \prime }=T^{*}(z)\) and \(u\in E(A, d^{\prime })\cap E(B, d^{\prime \prime })\). Therefore, we have

$$ S^{A\cap B}\left( T^{*}(z)\right)=u\cdot T^{*}(z)=u\cdot d^{\prime}+u\cdot d^{\prime\prime}=S^{A}(d^{\prime})+S^{B}(d^{\prime\prime}). $$
(2)

Equations (1) and (2) then imply that

$$ S^{A\cap B}(T^{*}(z))= \underset{d^{\prime},d^{\prime\prime} \in \ker(T)^{\perp}}{\underset{d^{\prime}+d^{\prime\prime}=T^{*}(z)}{\inf}} (S^{A}(d^{\prime})+S^{B}(d^{\prime\prime})). $$
(3)

As \(d^{\prime },d^{\prime \prime }\in \ker (T)^{\perp }=\text {image}(T^{*})\) there exist \(z^{\prime }, z^{\prime \prime }\in \mathbb {R}^{m}\) such that \(d^{\prime }=T^{*}(z^{\prime })\), \(d^{\prime \prime }=T^{*}(z^{\prime \prime })\), and \(z^{\prime }+z^{\prime \prime }=z\).

Hence,

$$ \begin{array}{@{}rcl@{}} S^{A\cap B}(T^{*}(z))&=&\inf_{{z^{\prime}+z^{\prime\prime}=z} \atop {z^{\prime},z^{\prime\prime} \in\mathbb{R}^{m} }} (S^{A}(T^{*}(z^{\prime}))+S^{B}(T^{*}(z^{\prime\prime})))\\ &=& \underset{z^{\prime},z^{\prime\prime} \in \mathbb{R}^{m} }{\underset{z^{\prime}+z^{\prime\prime}=z}{\inf}} (S^{TA}(z^{\prime})+S^{TB}(z^{\prime\prime}))=S^{TA\cap TB}(z). \end{array} $$

Overall, ST(AB)(z) = STATB(z) for all directions \(z\in \mathbb {R}^{m}\). Hence, T(AB) = TATB.

(Only-if statement) To establish the necessity part, we assume that T(AB) = TATB and consider any direction \(d\in \ker (T)^{\perp }\) and support point uE(AB, d). As \(\ker (T)^{\perp }=\text {image}(T^{*})\) there must exist \(z\in \mathbb {R}^{m}\) such that d = T(z). For this direction, we have

$$ S^{TA\cap TB}(z)=\underset{z^{\prime},z^{\prime\prime} \in \mathbb{R}^{m}}{\underset{z^{\prime}+z^{\prime\prime}=z}{\inf}} \left( S^{TA}(z^{\prime})+S^{TB}(z^{\prime\prime})\right)=\underset{z^{\prime},z^{\prime\prime} \in \mathbb{R}^{m} }{\underset{z^{\prime}+z^{\prime\prime}=z}{\inf}} \left( S^{A}(T^{*}(z^{\prime}))+S^{B}(T^{*}(z^{\prime\prime}))\right). $$
(4)

As T(AB) = TATB there must exist \(z^{\prime }, z^{\prime \prime }\in \mathbb {R}^{m}\) such that \(z^{\prime }+z^{\prime \prime }=z\), and

$$ u\cdot d=S^{A\cap B}(T^{*}(z))=S^{T(A\cap B)}(z)=S^{TA\cap TB}(z)=S^{A}(d^{\prime})+S^{B}(d^{\prime\prime}), $$

where \(d^{\prime }=T^{*}(z^{\prime })\), \(d^{\prime \prime }=T^{*}(z^{\prime \prime })\), and the last equality follows from Corollary 16.4.1 in [17], which asserts that the infimum of (4) is achieved when the relative interiors of the two sets have a point in common. Also, as uAB, we have \(u\cdot d^{\prime }\leq S^{A}(d^{\prime })\) and \(u\cdot d^{\prime \prime }\leq S^{B}(d^{\prime \prime })\). As a result, we must have \( u\cdot d^{\prime }=S^{A}(d^{\prime }), ~\text {and}~u\cdot d^{\prime \prime }=S^{B}(d^{\prime \prime }). \) In other words, \(u\in E(A, d^{\prime })\cap E(B, d^{\prime \prime })\), where \(d^{\prime }, d^{\prime \prime }\in \ker (T)^{\perp }\) and \(d^{\prime }+d^{\prime \prime }=d=T^{*}(z)\). As the choice of z is arbitrary and \(\text {image}(T^{*})=\ker (T)^{\perp }\) the only-if statement follows. □

Proof of Theorem 10

We want to prove that under the assumptions of the theorem, \(\overline {T(A\cap B)}=\overline {TA\cap TB}\) if and only if for any dker(T),

$$ \underset{d^{\prime},d^{\prime\prime} \in \mathbb{R}^{n} }{\underset{d^{\prime}+d^{\prime\prime}=d}{\inf}} (S^{A}(d^{\prime})+S^{B}(d^{\prime\prime})) = \underset{d^{\prime},d^{\prime\prime} \in \ker(T)^{\perp}}{\underset{d^{\prime}+d^{\prime\prime}=d}{\inf}} (S^{A}(d^{\prime})+S^{B}(d^{\prime\prime})). $$

The argument is similar to the proof of Theorem 9. Let us first consider the sufficiency part. For arbitrary direction \(z\in \mathbb {R}^{m}\), we have

$$ S^{T(A\cap B)}(z)=\sup_{t\in T(A\cap B)} t\cdot z=\sup_{x\in A\cap B} x\cdot T^{*}(z)=S^{A\cap B}(T^{*}(z)), $$

Given that \(T^{*}(z)\in \ker (T)^{\perp }\), the condition of the theorem implies

$$ S^{A\cap B}(T^{*}(z))=\underset{d^{\prime},d^{\prime\prime} \in \mathbb{R}^{n}}{\underset{d^{\prime}+d^{\prime\prime}=T^{*}(z)}{\inf}} (S^{A}(d^{\prime})+S^{B}(d^{\prime\prime}))=\underset{d^{\prime},d^{\prime\prime} \in \ker(T)^{\perp} }{\underset{d^{\prime}+d^{\prime\prime}=T^{*}(z)}{\inf}} (S^{A}(d^{\prime})+S^{B}(d^{\prime\prime})). $$

Similar to the proof of Theorem 9, as \(d^{\prime },d^{\prime \prime }\in \ker (T)^{\perp }=\text {image}(T^{*})\) there exist \(z^{\prime }, z^{\prime \prime }\in \mathbb {R}^{m}\) such that \(d^{\prime }=T^{*}(z^{\prime })\), \(d^{\prime \prime }=T^{*}(z^{\prime \prime })\), and \(z^{\prime }+z^{\prime \prime }=z\). Hence,

$$ \begin{array}{@{}rcl@{}} S^{A\cap B}(T^{*}(z))&=&\underset{z^{\prime},z^{\prime\prime} \in \mathbb{R}^{m} }{\underset{z^{\prime}+z^{\prime\prime}=z}{\inf}} (S^{A}(T^{*}(z^{\prime}))+S^{B}(T^{*}(z^{\prime\prime})))\\ &=&\underset{z^{\prime},z^{\prime\prime} \in \mathbb{R}^{m} }{\underset{z^{\prime}+z^{\prime\prime}=z}{\inf}} (S^{TA}(z^{\prime})+S^{TB}(z^{\prime\prime}))=S^{TA\cap TB}(z). \end{array} $$

Overall, ST(AB)(z) = STATB(z) for all directions \(z\in \mathbb {R}^{m}\). In contrast to Theorem 9, since T(AB) and TATB are not necessary closed, this only implies that \(\overline {T(A\cap B)}=\overline {TA\cap TB}\).

Let us now establish the necessity part. Assume that \(\overline {T(A\cap B)}=\overline {TA \cap TB}\). This implies that for any \(z\in \mathbb {R}^{m}\) we have ST(AB)(z) = STATB(z). We know that

$$ \begin{array}{@{}rcl@{}} S^{TA\cap TB}(z)&=&\underset{z^{\prime},z^{\prime\prime} \in \mathbb{R}^{m} }{\underset{z^{\prime}+z^{\prime\prime}=z}{\inf}} \left( S^{TA}(z^{\prime})+S^{TB}(z^{\prime\prime})\right)\\ &=&\underset{z^{\prime},z^{\prime\prime} \in \mathbb{R}^{m} }{\underset{z^{\prime}+z^{\prime\prime}=z}{\inf}} \left( S^{A}(T^{*}(z^{\prime}))+S^{B}(T^{*}(z^{\prime\prime}))\right). \end{array} $$

As \(\text {image}(T^{*})=\ker (T)^{\perp }\), there must exist \(d^{\prime }, d^{\prime \prime }\in \ker (T)^{\perp }\) such that \(d^{\prime }=T^{*}(z^{\prime }), d^{\prime \prime }=T^{*}(z^{\prime \prime })\). Since T is linear and bijective we also have that sets \(\{d^{\prime },d^{\prime \prime }\in ker(T)^{\perp } | d^{\prime }+d^{\prime \prime }=T^{*}(z)\}\) and \(\{d^{\prime },d^{\prime \prime }\in ker(T)^{\perp } | d^{\prime }=T^{*}(z^{\prime }), d^{\prime \prime }=T^{*}(z^{\prime \prime }), z^{\prime }+z^{\prime \prime }=z\}\) coincide. Therefore,

$$ S^{TA\cap TB}(z)=\underset{z^{\prime},z^{\prime\prime} \in \ker(T)^{\perp} }{\underset{z^{\prime}+z^{\prime\prime}=T^{*}(z)}{\inf}}\left( S^{A}(z^{\prime})+S^{B}(z^{\prime\prime})\right). $$

At the same time, we have

$$ S^{T(A\cap B)}(z)=S^{A\cap B}(T^{*}(z))=\underset{z^{\prime},z^{\prime\prime} \in \mathbb{R}^{n} }{\underset{z^{\prime}+z^{\prime\prime}=T^{*}(z)}{\inf}} \left( S^{A}(z^{\prime})+S^{B}(z^{\prime\prime})\right). $$

Finally, as image(T) = ker(T), we can conclude that for any dker(T),

$$ \underset{z^{\prime},z^{\prime\prime} \in \mathbb{R}^{n} }{\underset{z^{\prime}+z^{\prime\prime}=z}{\inf}} (S^{A}(z^{\prime})+S^{B}(z^{\prime\prime})) = \underset{z^{\prime},z^{\prime\prime} \in \ker(T)^{\perp} }{\underset{z^{\prime}+z^{\prime\prime}=z}{\inf}} (S^{A}(z^{\prime})+S^{B}(z^{\prime\prime})). $$

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Kushnir, A., Liu, S. On Linear Transformations of Intersections. Set-Valued Var. Anal 28, 475–489 (2020). https://doi.org/10.1007/s11228-019-00525-0

Download citation

Keywords

  • Linear transformation
  • Convex closed set
  • Intersection
  • Directional convexity
  • Mechanism design
  • Dominant-strategy implementation
  • Bayesian implementation

Mathematics Subject Classification (2010)

  • 52A20
  • 52A35
  • 91A99