行政院國家科學委員會專題研究計畫 成果報告
熵正則化及無窮多個比值的廣義型分數規劃問題
計畫類別: 個別型計畫 計畫編號: NSC91-2115-M-006-017- 執行期間: 91 年 08 月 01 日至 92 年 10 月 31 日 執行單位: 國立成功大學數學系暨應用數學所 計畫主持人: 許瑞麟 報告類型: 精簡報告 報告附件: 出席國際會議研究心得報告及發表論文 處理方式: 本計畫可公開查詢中 華 民 國 92 年 10 月 29 日
æñ
:G£†“£Ì¤Ö_ªMí2}bd•½æ
…û˝l•3bû˝©/|ü|×}b½æ
:(P) min
u∈Wmaxt∈T ft(u)
íÜ£l¶}
.憦}
,BbAŠíô.7
Crouzeix, Ferland andSchaible [1]
íj¶V°j½æ
(P).6ÿuz
,Bb„p7Ê
X¸
TÑ'KÕ¯
v
,†½æ
(P)í|7M}u-ƒbí;
(Pα) P (α) = min
x∈Xmaxt∈T [ft(x)− αgt(x)].
1/6°vz
Crouzeix, Ferland and Schaible [1]FZ|Víƶ ô.ƒ°
j©/|ü|×}b½æ
.7Ê¥ší²¬˙2
,â-Þí¥WVz
p
Tu'KÕ¯í½b4
.Example: Let X = [1/2, 1]× [1/2, 1] and T = [0, ∞).
5?}bd•
: inf (x1,x2)∈X sup t∈T{(−2x1− 1)/(tx1+ x2+ 1)}.%âø<lªJ)øw|7MÑ
α∗ = 0.çBb5?-Þí½æv
P (α) = inf (x1,x2)∈X sup t∈T{−2x1− 1 − α(tx1+ x2+ 1)},BbªJ)ø
úk
α > 0, P (α) = −2α − 3.úk
α = 0, P (α) = −3.úk
α < 0, P (α) = inf (x1,x2)∈X∞ = ∞.Ê¥_Wä2
,BbªJ)ø
P (α∗) = P (0) =−3.péD̪Mí|ü
|ד½æ!.¯
.-Bb„pç
X¸
T·u'KÕ¯v
,†ªJAŠ
à°j
P (α)í;V×)½æ
Pí|7M
.(1) Theorem If both X and T are compact, then P (α) < 0 if and only if α > α∗.
(2) Theorem If X and T are compact, then P (α∗) = 0. Moreover, problem (P)
and (Pα∗) have the same optimal solutions, that is, argmin
x∈X maxt∈T {ft(x)/gt(x)} = argminx∈X maxt∈T {ft(x)− α ∗g
t(x)}.
Ê¥_ô.|Víƶ2
,BbÊq¶c˛2Ûb5?ƒ°jø_©/
|ü|ד½æ
,¥³BbUà
Iterative Entropic Regularization [2]V¸
Dinkelbachƶ!¯
.yªø¥
,â¥_j¶BbZ|ø_ƶ
,Ê©ø_q¶c˛2.Ûb°jƒ|7jOuEÍUcñƶ)ƒY¹4
.à¤
,ªJªêrl|q¶c˛ í|7M)ƒyßíl^0
.¥_ƶ
Bb˚5Ñ
Modified Dinkelbach algorithm with entropic regularization,˚Ñ
MDER.2 Algorithm (MDER)
Step 1. Given x0∈ X, t1 ∈ T and η, γ > 0, 0 < δ < 1. Compute ˜
α1 = max
t∈T {ft(x0)/gt(x0)}.
Set p = 10, m = 1, T1={t1}, k = 1, j = 1 and ˜xmp, ˜α
k = x0.
Step 2. Approximate the following continuous minimax problem (P˜αk) P (˜αk) = min
x∈Xmaxt∈T [ft(x)− ˜αkgt(x)].
by the iterative entropic regularization [2]
Fp, ˜αm k(x) = (1/p) ln m i=1 exp(p(fti(x)− ˜αkgti(x))) . (1) Find ˜xmp, ˜α k that satisfies Fp, ˜αm k(˜xmp, ˜αk)≤ min x∈XF m p, ˜αk(x) + δj (2) and compute ˜ rk = max t∈T {ft(˜x m p, ˜αk)− ˜αkgt(˜xmp, ˜α k)}. (3)
Step 3. If ˜rk < −η, update the Dinkelbach parameter ˜ αk+1 = max t∈T {ft(˜x m p, ˜αk)/gt(˜xmp, ˜α k)}; (4)
the iteration counter k = k + 1; and η = η∗ (1 + δ). Go to Step 2.
Step 4. If ˜rk ≥ −η, set η = |˜rk|; improve m and p to obtain a better approximate solution. Step 4.1. If ˜rk > Fp, ˜αm k(˜xmp, ˜αk) + δk, choose tm+1 ∈ argmax t∈T {ft(˜x m p, ˜αk)− ˜αkgt(˜xmp, ˜αk)}. (5)
Set Tm+1= Tm∪ {tm+1}, p = p1+ε and j = j + 1. Go to Step 2. Step 4.2. δj+ (ln m)/p > γ− δk, set p = p1+ε and j = j + 1. Go to Step 2. Step 5. (Stopping Criteria)
If η > γ, compute ˜αk+1 = max
t∈T {ft(˜x m
p, ˜αk)/gt(˜xmp, ˜α
k)}, k=k+1 and return to
Step 2. Otherwise, stop and report that ˜xmp, ˜α
k is a γ-optimal solution of (P).
úk¥_ƶ
,Bb)ƒø<½bíùÜ£ìÜà-
,¥<uàV„p
MDER
íY¹45à
(3) Lemma Let α∗ be the optimal value of problem (P). Then, ˜αk ≥ α∗.
(4) Lemma The optimal value P ( ˜αk) of parametric problem P˜αk is non-positive.
(5) Lemma Fp, ˜αm
k(x∗˜αk)≤ maxt∈Tm[ft(x∗˜αk)− ˜αkgt(x∗˜αk)] + (ln m)/p
(6) Lemma If ˜rk ≤ Fp, ˜αm
k(˜xmp, ˜αk) + δk, then
(7) Lemma If max{−η, P (˜αk)} ≤ ˜rk ≤ P (˜αk) + (ln m)/p + δj+ δk, then −δk− δj− (ln m)/p − η ≤ P (˜αk)≤ 0.
-¥_ìÜuzpçƶÊ
Step5¢ÏWv
, P ( ˜αk)¸
0íÏÏBÖ
u
2γ,6ÿuBbªJ)ƒø_¸ˇÊ
2γí¡Nj
(2γ-optimal solution).(8) Theorem When the algorithm stops in Step 5 with k = ¯k, then the following inequality
−2γ ≤ P (˜α¯k)≤ 0
hold.
(9) Lemma If the algorithm does not stop at Step 5, then
˜
rk < −γ < 0 and
−γ + ˜rk ≤ P (˜αk).
(10) Lemma {˜αk} is a decreasing sequence.
(11) Lemma If ˜αk > ˜αk+1, then P ( ˜αk)≤ P (˜αk+1) + ( ˜αk+1− ˜αk) min
t∈T gt(x ∗
˜αk+1).
(12) Lemma If ˜αk > α∗, then P ( ˜αk)≤ P (α∗) + (α∗− ˜αk) min
t∈T gt(x ∗). (13) Theorem If P ( ˜αk) < ˜rk < −γ < 0, then ( ˜αk+1− α∗)/( ˜αk− α∗)≤ 1 − m(g)/M(g) − ˜rk/( ˜αk − α∗) max t∈T gt(˜x m p, ˜αk) .
(14) Condition There exists some positive integer K such that
0 < εk < (m(g)/M(g)) − γ, ∀k > K.
°v
,Ê_çí‘K-
(Condition 14),BbªJ„p
MDERíY¹§u
(4í
.(15) Theorem Under Condition 14, if the algorithm does not stop at Step 5, then, from certain iterate K on,
( ˜αk+1− α∗)/( ˜αk − α∗)≤ 1 − γ, ∀k > K
In other words, {˜αk} converges to α∗ linearly.
Êl,
,BbõÒÏW7ù_Wä
,Example1.
¥uø_(4}bd•
min
x∈Xmaxt∈T {(−tx1+ x2)/(x1+ x2)}
w2
X = [1/2, 30] × [1/2, 100]¸
T = [0, 1].¥_Wäí|7MªJ%âøOí
l)ƒu
1/61,|7õ†êÞÊ
(30, 1/2).Ê
Table 1.Bb|7˙FUà
í¡b£l!‹
.Table 1 Computational results for Example 1
x0 δ γ Iters. x∗ Optvalue Time
(1/2, 100) 0.25 1e-5 38 (30, 0.5) 0.016393 2.016
â¥WäBbªJ)ƒí!u¥_ƶªJAŠíj|
Example 1í
4 Example 2.
ù_Wäu-í}bd•
min x∈Xmaxt∈T {(t 2x 1x2+ x2t1 + tx33)/(5(t− 1)2x41+ 2x22+ 4tx3)}w2
(x1, x2, x3)∈ X = [1/2, 5] × [1/2, 5] × [1/2, 5]¸
t ∈ T = [0, 1].ƶ
MDERà
(5, 5, 5)ç–áõ1/!!Ê
(0.5, 1.4916, 0.5),7|7MÑ
−0.209954.Bbà
P (−0.209954) = min x∈Xmaxt∈T {f(t) + 0.209954g(t)} = 0£
Theorem 1¸
2‡ì
(0.5, 1.4916, 0.5)íüu
Example 2í|7j
.¥<!‹
Bb}Ê
Table 2¸
Table 3.Table 2 Computational results for Example 2
x0 γ Iters. x∗ Optimal value Time
(5,5,5) 1e-2 112 (0.5, 1.491666160, 0.5) -0.209954624 5.703 Table 3 Verification of optimal value -0.209954 by Matlab
α P (α)
-0.209554 -3.4010e-005
*
Table 5.42ªJõ|
MDERÊLSøjÞ̬f$í
Dinkelbachal-gorithm,
w2¨7ÏWvÈ
,LHŸbJ£jíü
¡5d.
[1] J. P. Crouzeix, J. A. Ferland, and S. Schaible. An algorithm for generalized fractional programs.
Journal of Optimization Theory and Applications 47 (1985), 35–49.
[2] R. L. Sheu and J. Y. Lin. On solving continuous min-max problems by iteratively entropic-regularization method. To appear in Journal of Optimization Theory and Applications (2003).