何毓琦的个人博客分享 http://blog.sciencenet.cn/u/何毓琦 哈佛(1961-2001) 清华(2001-date)

博文

How to Do Research #3 精选

已有 18715 次阅读 2007-5-28 03:53 |系统分类:科研笔记

(For new reader and those who request 好友请求, please read my 公告栏 first)

HOW TO DO RESEARCH 3 – PRACTICE WHAT YOU PREACH

I am most gratified by the response to my two blog articles "How to Do Research?" and
"More on How to Do Research". In the space of one week a total of 1500+ persons have
read the articles. In the spirit of practicing what I preach, but at the risk of self-promotion,
let me give you an example on how my recommendation on "finding a good research
topic" benefited myself. For convenience I repeated the recommendation here:

“GO FIND A REAL WORLD PROBLEM THAT A GROUP OF PEOPLE IS EAGER
TO SOLVE, THAT HAPPENS TO INTEREST YOU FOR WHATEVER REASON,
AND THAT YOU DON'T KNOW MUCH ABOUT. MAKE A COMMITMENT TO
SOLVE IT BUT NOT A COMMITMENT TO USE TOOLS WITH WHICH YOU
HAPPEN TO BE FAMILIAR" (But you should really read this rule in the context of the
whole article)



I.THE REAL WORLD PROBLEM:

Civilization have increasingly created complex human-made systems, such as large-scale
electric power grids, air and land traffic control systems, manufacturing plants and supply
chains, the Internet and other communication networks, etc. Such systems operate and
evolve in time via human made rules of operation, and are difficult to describe and
capture by succinct mathematical models such as differential equations for physical
systems. Collectively, they are known as Discrete Event Dynamic Systems (DEDS). In
fact, the only faithful description of such complex systems is by way of an electronic
copy, i.e., a simulation model/program that duplicates everything the real system does in
real or simulation time. Evaluation of the performance of such system is accomplished by
running such simulation models rather than experimenting with the real systems. To give
a familiar example, Consider the Beijing Capital International Airport. On the landside,
there are passengers, taxis, buses, and private cars. On the airside, there are planes,
military, commercial , and private. All these entities arrive stochastically or are subject to
stochastic disturbance such as weather or equipment breakdown. They all require and
compete for a series of services from facilities at the airport. For example, in the case of a
passenger arriving by car, s/he need parking spaces, luggage handling, check in, security
inspection, rest room, snack place, and waiting lounges. It is easy to see that such a
system can only be modeled accurately via a simulation model. And simulating and
experimenting with the behavior of such a system over a period of say 24 hours, will
require considerable amount of computer time. Typically, the time required for one
particular evaluation of the system performance can take hours or even days of computer
time.
But this is not the end of the problem difficulty. Other fundamental mathematical
limitations further complicates the problem.

It is a well known statistical fact that under the best of circumstance the accuracy of an
estimate improves very slowly with the number of samples used to calculate the
estimate.. For every one order of magnitude increase in accuracy of the estimate, the
number of samples must increase by two orders of magnitude. Since each sample run of
the simulation model of a complex system may consume considerable time, running the
simulation model many times to achieve an accurate estimate of the performance may
impose a heavy burden. And if optimization using the system designs parameters is
planned, the total amount of computation required quickly becomes infeasible. As a
result, simulation model are often used for validation of a design obtained by other means
and not for optimization purposes.

To make matters worse, while the literature on optimization and decision-making is huge,
much of the concrete analytical results are associated with what may be called Real
Variable Based methods. The idea of successive approximation to an optimum (say,
minimum) by sequential improvements based on local information is often captured by
the metaphor of "skiing downhill in a fog". The concepts of gradient (slope), curvature
(valley), and trajectories of steepest descent (fall line) all require the notion of derivatives
and are based on the existence of a more or less smooth response surface. There exist
various first and second order algorithms of feasible directions for the iterative
determination of the optimum (minimum) of an arbitrary multi-dimensional response or.
performance surface. Considerable numbers of major success stories exist in this genre
including the Nobel Prize winning work on linear programming. It is not necessary to
repeat or even reference these here.

On the other hand, in spite of the tremendous development of the science and art of
optimization and computation, there remain many problems that are still beyond our
reach. Among them are the class of combinatorial NP-hard problems and the well known
"curse of dimensionality" in dynamic programming. Exponential growth is one law that
mathematics and computers cannot overcome.

Thus summarizing, the problems are:
oSYSTEM COMPLEXITY
oCOMPUTATIONAL INTENSITY
oLACK OF EXTANT LITERATURE
oFUNDAMENTAL MATHEMATICAL LIMITATIONS

II. IDEAS FOR SOLUTION

There are two basic ideas in the methodology of Ordinal Optimization (OO, "序优化") :
:
o"Order" is much more robust against noise than "Value", i.e. "Asking which is
better?" is a much easier question than "how much better"
oDon't insist on getting the "Best" but be willing to settle for the "Good Enough" –
remember the proverb "best is the enemy of good enough"

Space limitation prevent me to further elaborate on these simple notions by intuitive
explanations. More details can be found below in section III. Essentially we are making a
slight compromise in what we are asking in order to gain advances on the solution front.

Of course readers may rightly point out that these ideas are hardly new and very simple.
Good engineers and designers do this all the time when confronted with difficult and
complex problems of performance evaluation and optimization. My own contribution is
simply that we have developed a theory to QUANTIFY these two ideas. The practice of
these two ideas is now knowledge-based instead of being experience-based. The expertise
of a good designer acquired from experience will now be available to everyone who uses
the tools discussed in this methodology. Moreover, the user will have numerical measures
rather than just gut feelings.

III. APPLICATION EXAMPLES

There are now over 250 references on the topics of Ordinal Optimization (OO,
"序优化"). See the annotated list at cfins.au.tsinghua.edu.cn.  But my favorite example is
the one that led to the invention of OO which I discovered one rainy afternoon in 1991,
which I still use to demonstrate the two basic ideas, which you can set up yourself on
EXCEL in about 15 minutes, and which can also be found on my web site
www.hrl.harvard.edu/~ho. By September 2007, a complete book on this topic which I co-
authored with my Tsinghua colleagues will be published by Springer.

IV. THE METHODOLOGY OF ORDINAL OPTIMIZATION AND IT
RELATIONSHIP TO
OTHER TOOLS

"Optimization" taken in the broadest sense as seeking improvement is an idea as old as
the existence of human beings. In fact, it can be argued that it is the "raison d'être" for
our civilization. Without the desire to improve, progresses on all fronts will stall. Yet the
study of optimization as a discipline and not as individual endeavors on specific problems
did not begin until the invention of calculus, which enabled the mathematical modeling of
large number of physical phenomena. The theory of maxima/minima and convexity
emerged as a result. Yet the numbers of real world problems that can be explicitly solved
by mathematics alone remain limited until the development of the computer. Suddenly,
many algorithms, which previously thought to be infeasible for the numerical and
iterative solution of difficult optimization problems, now become possible. The golden
age of optimization took off in the 1950s and is still ongoing.
On the other hand, in spite of the tremendous development of the science and art of
optimization and computation, there remain many problems that are still beyond our
reach. Among them are the class of combinatorial NP-hard problems and the well known
"curse of dimensionality" in dynamic programming. Exponential growth is one law that
mathematics and computers cannot overcome. Furthermore, computational burden of a
problem does not always necessarily arise because of problem size. Complexity of a
problem can also impose infeasible computational burdens as the simulation model of
human made systems.
.
Furthermore, we submit that the reason many real world optimization problems remain
unsolved is partly due to the changing nature of the problem domain, which makes
calculus or real variable based method less applicable. For example, a large number of
human-made system problems mentioned above involve combinatorics, symbolic or
categorical variables rather than real analysis, discrete instead of continuous choices, and
synthesizing a configuration rather than proportioning the design parameters.
Optimization for such problem seems to call for general search of the performance terrain
or response surface as opposed to the "skiing downhill in a fog" metaphor of real variable
based performance optimization . Arguments for change can also be made on the
technological front. Sequential algorithms were often dictated as a result of the limited
memory and centralized control of earlier generations of computers. With the advent of
modern massively parallel machines, distributed and parallel procedures can work hand-
in-glove with Search Based method of performance evaluation. The purpose of OO is to
address the difficulties of optimization problems described above – the optimization of
complex systems via simulation models or other computation-intensive models involving
possible stochastic effects and discrete choices.

V. CONCLUSION

Of course, sections I-IV represent hindsight. When I started out on that rainy afternoon in
1991 or published the first paper in 1992, I certainly did not have this vision.
But I knew enough at the time that the ideas of OO addressed most of the difficulties of
the real world problems and had success on some computational issues previously
thought to be infeasible.  I also had three decades of experience with and confidence in
my own recommended rule of research  that I applied for funding, encouraged graduate
students and postdocs, and dedicated my own efforts for 15 years on the subject.. It is
what researchers called a first generation research topic (or what I denote as the second
level of research problem discussed in my first blog on the subject of "how to do
research".


Note added 5/11/2106: There is a nice Chinese summary of this and related article at http://blog.sciencenet.cn/home.php?mod=space&uid=535297&do=blog&id=636074  何毓琦院士教年轻人如何做科研 精选




https://blog.sciencenet.cn/blog-1565-2501.html

上一篇:小女儿毕业典礼
下一篇:中文的自我介绍及一个月来做博客的感触
收藏 IP: 74.104.133.*| 热度|

1 鲁雪松

该博文允许注册用户评论 请点击登录 评论 (5 个评论)

1/0 | 闂傚倸鍊搁崐鎼佸磹閹间礁纾归柟闂寸绾惧綊鏌熼梻瀵割槮缁炬儳缍婇弻鐔兼⒒鐎靛壊妲紒鐐劤缂嶅﹪寮婚悢鍏尖拻閻庨潧澹婂Σ顔剧磼閻愵剙鍔ゆ繝鈧柆宥呯劦妞ゆ帒鍊归崵鈧柣搴㈠嚬閸欏啫鐣峰畷鍥ь棜閻庯絻鍔嬪Ч妤呮⒑閸︻厼鍔嬮柛銊ョ秺瀹曟劙鎮欏顔藉瘜闂侀潧鐗嗗Λ妤冪箔閹烘挶浜滈柨鏂跨仢瀹撳棛鈧鍠楅悡锟犮€侀弮鍫濋唶闁绘柨寮剁€氬ジ姊绘担鍛婂暈缂佽鍊婚埀顒佸嚬閸o綁宕洪姀鈥崇窞闁归偊鍘鹃崢鍗炩攽閳藉棗鐏犻柣蹇旂箖缁傚秹宕烽鐘碉紲濡炪倖妫侀崑鎰櫠閿旈敮鍋撶憴鍕闁靛牊鎮傞獮鍐閵忋垻鐓撻梺鍓茬厛閸犳洜妲愰悢灏佹斀闁绘ɑ鍓氶崯蹇涙煕閻樻剚娈滈柕鍡楀暣瀹曘劑顢橀崶銊р槈閾绘牠鏌涘☉鍗炲箻妞わ富鍣e娲箰鎼淬垻顦ラ梺绋匡工缂嶅﹪骞冮敓鐘参ㄩ柨鏂垮⒔閻﹀牓姊婚崒姘卞缂佸甯¢弫宥咁吋閸℃洜绠氶梺鍦帛鐢骞夐崫銉х<閺夊牄鍔屽ù顕€鏌熼瑙勬珚闁诡喗绮岃灒闁绘挸瀛╅柨顓熺節閻㈤潧鍓崇紒鑼舵鐓ら柕鍫濐槹閺呮繃銇勮箛鎾村櫢缂佽妫濋弻娑㈩敃閿濆棛顦ュ┑锛勫仒缁瑩寮诲☉銏犵疀闁靛⿵闄勯悵鏇㈡⒑閸濆嫭顥欓柛妤€鍟块~蹇曠磼濡顎撻梺鍛婄☉閿曘儵宕曢幘缁樷拺鐟滅増甯楅弫閬嶆煕閵娿儲璐℃俊鍙夊姍閹瑧鈧稒锚椤庢捇姊洪崨濠冨碍鐎殿喖澧庣槐鐐存償閵婏腹鎷洪梺璇″瀻閸涱垼鍟堟俊鐐€ら崑鍕囬鐐村仼闂佸灝顑呯欢鐐烘煙闁箑骞橀柛姗嗕簼缁绘繈鎮介棃娑楃捕闂佽绻戠换鍫濈暦濠靛棌鍫柛顐ゅ枔閸樿棄鈹戦悩缁樻锭閻庢凹鍓熼幃姗€宕f径瀣伎婵犵數濮撮崯顖炲Φ濠靛鐓欐い鏃€鍎抽崢瀵糕偓娈垮枛閻栧ジ鐛幇顓熷劅妞ゆ柨鍚嬮弳蹇涙⒒閸屾艾鈧兘鎳楅崼鏇炵疇闁规崘顕ч崥褰掓煛瀹ュ骸骞栫紒鐙€鍨堕弻銊╂偆閸屾稑顏�:0 | 婵犵數濮烽弫鍛婃叏閻戣棄鏋侀柛娑橈攻閸欏繘鏌i幋锝嗩棄闁哄绶氶弻娑樷槈濮楀牊鏁鹃梺鍛婄懃缁绘﹢寮婚敐澶婄婵犲灚鍔栫紞妤呮⒑闁偛鑻晶顕€鏌涙繝鍌涘仴妤犵偞鍔栫换婵嬪礃椤忓棗楠勯梻浣稿暱閹碱偊顢栭崶鈺冪煋妞ゆ棃鏁崑鎾舵喆閸曨剛锛橀梺鍛婃⒐閸ㄧ敻顢氶敐澶婇唶闁哄洨鍋熼娲⒑缂佹ḿ鎳冮柟铏姍閻涱噣濮€閵堝棌鎷婚梺绋挎湰閻燂妇绮婇悧鍫涗簻妞ゆ劑鍩勫Σ鎼佹煟閿濆懎妲婚摶锝夋偠濞戞帒澧查柣蹇撳暙閳规垿鎮欓弶鎴犱桓缂佺偓婢樼粔褰掋€侀弴鐔侯浄閻庯綆鍋嗛崢顏堟⒑閸撴彃浜濈紒璇插暣瀹曨垶骞掑Δ浣哄幐闂佸憡鍔戦崝搴㈡櫠濞戙垺鐓涢柛娑卞枤閸欌偓闂佸搫鏈粙鎴﹀煡婢舵劕纭€闁绘劘灏欓鎴炵節閻㈤潧浠滈柣妤€锕幃锟犲灳閹颁焦缍庨梺鎯х箰濠€閬嶆儗濞嗘劗绠鹃柛鈩兠崝銈夋煕閹惧瓨绶叉い顏勫暣婵″爼宕卞Δ鍐ф樊婵$偑鍊х粻鎾翅缚瑜旈、姘舵晲閸℃瑧鐦堝┑顔斤供閸樿棄鈻嶅⿰鍫熲拺闁告稑锕﹂埥澶愭煕婵犲偆鐓肩€规洜澧楅幆鏃堝Ω閵壯冨箳闂佺懓鍚嬮悾顏堝礉瀹€鈧划璇差潩鏉堛劌鏋戦柟鍏兼儗閻撳牓寮繝鍥ㄧ厱闁哄洢鍔岄悘锟犳煟閹惧鈽夋い顓℃硶閹瑰嫰鎮滃鍡橈紒婵犵數鍋涢幊搴∥涘☉姘潟闁圭儤姊圭€氭岸鏌ょ喊鍗炲妞わ絽鎼—鍐Χ鎼粹€茬盎缂備胶绮敃銏ょ嵁閺嶎厼鎹舵い鎾跺枎閺嬪倿姊洪崨濠冨闁稿妫濋弫宥堢疀濞戞瑢鎷绘繛鎾村焹閸嬫挻绻涙担鍐插娴犳岸姊绘担鍛靛湱鈧稈鏅犻幃锟犳晸閻樿尙鐣洪梺姹囧灮鏋い顐㈡嚇閺屾洟宕煎┑鍥舵闂佸綊鏀卞钘夘潖濞差亝鍤掗柕鍫濇噺閻庢儳鈹戦悩顔肩仾闁挎岸鏌嶇紒妯诲磳闁糕晪绻濆畷銊╊敊閹冪闂傚倷绀侀幉锟犲垂椤栫偛纾归柡宥庡亐閸嬫挸顫濋悙顒€顏� | 婵犵數濮烽弫鍛婃叏閻戣棄鏋侀柛娑橈攻閸欏繘鏌i幋锝嗩棄闁哄绶氶弻娑樷槈濮楀牊鏁鹃梺鍛婄懃缁绘﹢寮婚敐澶婄闁挎繂妫Λ鍕⒑閸濆嫷鍎庣紒鑸靛哺瀵鎮㈤崗灏栨嫽闁诲酣娼ф竟濠偽i鍓х<闁绘劦鍓欓崝銈嗐亜椤撶姴鍘寸€殿喖顭烽弫鎰緞婵犲嫮鏉告俊鐐€栫敮濠囨倿閿曞倸纾块柟鍓х帛閳锋垿鏌熼懖鈺佷粶濠碘€炽偢閺屾稒绻濋崒娑樹淮閻庢鍠涢褔鍩ユ径鎰潊闁冲搫鍊瑰▍鍥⒒娴g懓顕滅紒璇插€歌灋婵炴垟鎳為崶顒€唯鐟滃繒澹曢挊澹濆綊鏁愰崨顔藉創閻忓繐绻樺娲川婵犲孩鐣锋繝鐢靛仜閿曘倝顢氶敐鍡欑瘈婵﹩鍘兼禍婊堟⒑缁嬭法绠伴柣銊у厴楠炲繑绻濆顓犲幗闁硅壈鎻徊楣冨汲閳哄懎鍑犳い蹇撳閺€浠嬫煃閵夈劌鐨洪柣顓熺懄閹便劍绻濋崘鈹夸虎閻庤娲﹂崑濠傜暦閻旂⒈鏁嗛柍褜鍓欓埢宥夊Χ閸モ晝锛濇繛杈剧悼閻℃棃宕靛▎鎾寸厽婵°倓鐒︾亸顓熴亜閺囶亞绋荤紒缁樼箓椤繈顢栭埞鐐珚闁哄本娲樺鍕醇濠靛牅鐥梻浣告惈閸婄敻宕戦幘缁樷拻闁稿本鐟ㄩ崗宀€鐥鐐靛煟鐎规洘绮岄埞鎴犫偓锝庡亝濞呮梻绱撻崒姘偓鐑芥倿閿曞倸绀夐柡宥庡幗閸庡孩銇勯弽銊ュ毈婵炲吋鐗楃换娑橆啅椤旇崵鐩庨悗鐟版啞缁诲倿鍩為幋锔藉亹闁圭粯甯╅崝澶愭⒑娴兼瑧鎮奸柛蹇旓耿楠炲啫螖閸涱厾顦ф繝銏f硾閿曪絾绔熼弴銏♀拻濞达絽鎽滅粔鐑樹繆椤愩儲纭剁紒顔肩墛閹峰懘鎼归柅娑氱憹闂備礁鎼粔鏌ュ礉鐎n剚宕查柛鈩冪⊕閻撳繘鏌涢锝囩畵闁逞屽墮閹诧紕绮嬪鍡愬亝闁告劏鏂侀幏娲煟鎼粹剝璐″┑顔炬暬钘熷璺侯儍娴滄粓鏌ㄩ弮鍥跺殭闁诲骏绠撻弻娑㈠煘閸喚浠煎銈嗘尭閵堢ǹ鐣烽崡鐐嶇喖鎳栭埞顑惧€濆缁樼瑹閳ь剙岣胯閸e綊姊洪崨濠佺繁闁搞劍澹嗛弫顕€骞掗弮鍌滐紳闂佺ǹ鏈懝楣冨焵椤掑嫷妫戠紒顔肩墛缁楃喖鍩€椤掆偓閻g兘骞囬弶澶哥炊闂侀潧锛忛崨顖氬脯闂傚倷绀佸﹢閬嶆惞鎼淬劌绐楅柟鎹愵嚙绾惧鏌熺€涙ḿ璐╃憸鐗堝笒缁€鍌炴煕韫囨艾浜圭紒瀣喘濮婄粯鎷呯粙鑳煘濠电偛妯婇崣鍐嚕婵犳碍鏅插璺猴攻椤ユ繈姊洪崷顓€鍦偓娑掓櫊瀹曟洟骞樼紒妯衡偓鍨箾閸繄浠㈤柡瀣枎閳规垿鎮欑拠褍浼愬銈庡亜缁绘帞妲愰幒鎳崇喓鎷犲顔瑰亾閹剧粯鈷戦柟顖嗗懐顔婇梺纭呮珪閹稿墽鍒掗銏犵伋闁哄倶鍎查弬鈧梻浣虹帛閸旀牞銇愰崘顏嗘/鐟滄棃寮婚敐鍛傛棃宕橀妸鎰╁灲閺岋綁鏁愰崶褍骞嬪Δ鐘靛仜濞差厼顕i崼鏇炵閹艰揪绱曢妶璺衡攽閿涘嫬浜奸柛濠冪墵楠炴劙鎳¢妶鍥╃厯闂佺懓顕崑鐔笺€呴弻銉︾厽闁逛即娼ф晶顖炴煕濞嗗繒绠插ǎ鍥э躬椤㈡稑饪伴崘銊ょ帛濠电偛鐡ㄧ划鎾剁不閺嶎厼绠栨俊銈傚亾妞ゎ偅绻堥幃娆擃敆閳ь剟顢旈敓锟� | 闂傚倸鍊搁崐鎼佸磹閹间礁纾归柟闂寸绾惧綊鏌熼梻瀵割槮缁炬儳缍婇弻鐔兼⒒鐎靛壊妲紒鐐劤濠€閬嶅焵椤掑倹鍤€閻庢凹鍙冨畷宕囧鐎c劋姹楅梺鍦劋閸ㄥ綊宕愰悙鐑樺仭婵犲﹤鍟扮粻鑽も偓娈垮枟婵炲﹪寮崘顔肩<婵炴垶鑹鹃獮鍫熶繆閻愵亜鈧倝宕㈡禒瀣瀭闁割煈鍋嗛々鍙夌節闂堟侗鍎愰柣鎾存礃缁绘盯宕卞Δ鍐唺缂備胶濮撮…鐑藉蓟閳ュ磭鏆嗛柍褜鍓熷畷浼村箻閼告娼熼梺鍦劋椤ㄥ懘锝為崨瀛樼厽婵☆垵娅i敍宥吤瑰⿰搴濈敖缂佽鲸鎹囧畷鎺戔枎閹烘垵甯┑鐘愁問閸o絿绮婚弽顓熸櫜闁绘劖娼欑欢鐐烘煙闁箑鍔﹂柨鏇炲€归悡鏇㈡煛閸ャ儱濡奸柣蹇曞Х缁辨帡鎮╁畷鍥ㄥ垱闂佸搫鏈惄顖炪€侀弴銏℃櫜闁糕剝鐟Σ鐗堜繆閻愵亜鈧洘顨ラ崫銉х煋闁荤喖鍋婂ḿ鏍煣韫囨挻璐¢柣顓熺懄缁绘盯宕卞Ο鍝勫Б闂佸憡鎸鹃崑鎾舵崲濞戞埃鍋撳☉娆嬬細闁活厹鍊曢湁婵犲﹤绨肩花缁樸亜閺囶亞鎮奸柟椋庡Т闇夐悗锝庡亽濞兼棃姊绘笟鈧ḿ褏鎹㈤幒鎾村弿闁割偁鍎辨儫闂佹寧妫佸銊ц姳婵犳碍鈷戦柛婵嗗閳ь剚鎮傞幃妯衡攽閸垻顦梺鍝勭Р閸斿秹宕h箛娑欏仭婵炲棗绻愰瀛樼箾閸喓鐭掗柡宀€鍠栭、娆撴偂鎼粹懣鈺佄斿Δ濠佺胺闁告鍟块悾鐑藉Ω閳哄﹥鏅i悷婊冮琚欏鑸靛姈閳锋垶鎱ㄩ悷鐗堟悙闁绘帗妞介弻娑㈠Ω閳衡偓閹查箖鏌曢崱妤€鏆炵紒缁樼箞瀹曟帡濡堕崨顕呭悪闂傚倷绀侀幖顐ゆ偖椤愶箑纾块弶鍫氭櫇娑撳秹鏌i悢绋挎珵鐟滅増甯楅弲鏌ユ煕濞戝崬鏋︾痪顓涘亾闂傚倷绀侀幉锟犳偡閵夈儙娑樷攽閸♀晜缍庨梺鎯х箰濠€杈╁閸忛棿绻嗘い鏍ㄧ箓閸氳銇勯敂鍝勫姦婵﹨娅g划娆撳礌閳ュ厖绱f繝鐢靛Л閸嬫捇姊洪鈧粔鎾倿閸偁浜滈柟鍝勭Х閸忓矂鏌涢悢鍝ュ弨闁哄瞼鍠栧畷娆撳Χ閸℃浼�

扫一扫,分享此博文

Archiver|手机版|科学网 ( 京ICP备07017567号-12 )

GMT+8, 2025-3-18 19:46

Powered by ScienceNet.cn

Copyright © 2007-2025 中国科学报社

返回顶部