One of the surprising aspects of this year has been the astonishing lack of performance from Williams and McLaren - and, at the beginning of the season, Force India.
People often wonder how, with the incredible array of technology available to the teams, performance drop-offs like this are possible. They're not just confined to the lesser-funded teams; in 2015 Red Bull produced a car that didn't begin to show the team's normal potential until mid-season.
So what can be the reason? The first thing to remember is that a Formula 1 car needs to be thought of in system engineering terms.
Performance isn't a function of any single aspect of those inter-related systems. Wise choices at the design stage, together with a full understanding of the influence of certain parameters on lap time and straightline speed at different circuits, pay dividends throughout the season.
While all design variables influence each other, there are certainly some that have a greater connection to overall performance than others. The three most influential over a range of circuits are power, grip and aerodynamic efficiency. The first two, to some extent, can be compared across cars. There are only four types of power unit and, with the exception of Honda's, they are used in different chassis. This can rule out gross differences in performance in this area. There will always be some differences associated with the installation of the power unit in the car, but these are very much second-order influencers.
Similarly, all the cars are using the same tyres, which gives a degree of performance normalisation - especially on a single qualifying lap on new tyres. While it's true that the current tyres are very sensitive to operating conditions, when analysed statistically over a number of events, a pattern should rapidly emerge.
This leaves aerodynamics as the most likely cause of disparity between expected and actual results. But surely the sophistication of the tools these days should put paid to that type of concern? Oh, that it were so.
The aerodynamics of an open-wheel racing car are immensely complex. Much of the flow is turbulent and this makes it more difficult to predict using Computational Fluid Dynamics (CFD), and even a windtunnel, since there are many aspects of the flow that are difficult to model.
Both CFD (in the form commonly used by teams) and windtunnel testing operate in a time-averaged domain. What this means is they effectively measure forces or conditions which will be varying over a period of time, and then take an average result over this time period. In isolation this can give a very poor picture of actual performance.
Let's take a simple case where the downforce is varying in a simple, relatively slowly repeating pattern, like a sine wave with a variation that goes both 10% above the average and 10% below. The aerodynamicist will report the average number but the driver will only be able to exploit something around the minimum downforce, which is maybe 8-9% below the average.
Now consider a car that has nominally the same downforce, in other words the average figure is the same, but now the variation is plus or minus 15% from that average.
The driver once again will drive to somewhere just above the minimum, say 13-14%, which will give them considerably less downforce than on the previous car, even though the reported number is the same. In fact it may be worse than this because the second car, having more variation, will undoubtably give the driver a feeling of severe instability, discouraging them from finding the limit.
In addition to these sort of problems, it is common to hear teams reporting that they have lost the correlation between their windtunnel and the car on the track. Most people assume this means that the aerodynamic measurements made in the windtunnel are different to those measured on the car. There is an infernal triangle consisting of CFD results, windtunnel results and car results, and none of them give a true answer. One might think that the car must be the best measurement but unfortunately a racing car isn't a scientific device, and by instrument standards the measurements are somewhat crude.
On top of all that, the aerodynamic forces experienced by the wheel, be they the lift of the rotating wheel or the downforce produced by the brake duct winglets, are difficult to measure because they aren't passing through the sprung mass.
So what is good correlation? The answer lies not just in being able to measure numbers that when fed into a simulation will give a reasonable prediction of lap time, but more importantly a situation whereby trends can be followed with confidence. If the flow field predicted by CFD or even that experienced in the windtunnel is somewhat different to that seen on the car, then it becomes extremely difficult to predict the effect of changes.
When Williams had a problem after fitting a new rear wing at Silverstone this year the problem didn't lie in the rear wing itself, but in the flow detaching from the floor when the DRS was used and then not re-attaching when the DRS closed again. Not only was this not detected in the windtunnel, it also wasn't detected during practice the previous day.
This is a measure of just how critical the aerodynamic performance can be to the conditions under which it performs. When I was at Williams we even used to clean the leading edge of the rear wing in pitstops because the dead flies it collected could adversely affect its performance.
Good correlation exists when any trend seen on the car is predicted by the experimental techniques used to develop the car. Easily said but not easily achieved, and the more complex and critical the flow fields are that deliver performance, the more difficult that nirvana of correlation becomes.
One of the surprising aspects of this year has been the astonishing lack of performance from Williams and McLaren - and, at the beginning of the season, Force India.
People often wonder how, with the incredible array of technology available to the teams, performance drop-offs like this are possible. They're not just confined to the lesser-funded teams; in 2015 Red Bull produced a car that didn't begin to show the team's normal potential until mid-season.
So what can be the reason? The first thing to remember is that a Formula 1 car needs to be thought of in system engineering terms.
Performance isn't a function of any single aspect of those inter-related systems. Wise choices at the design stage, together with a full understanding of the influence of certain parameters on lap time and straightline speed at different circuits, pay dividends throughout the season.
While all design variables influence each other, there are certainly some that have a greater connection to overall performance than others. The three most influential over a range of circuits are power, grip and aerodynamic efficiency. The first two, to some extent, can be compared across cars. There are only four types of power unit and, with the exception of Honda's, they are used in different chassis. This can rule out gross differences in performance in this area. There will always be some differences associated with the installation of the power unit in the car, but these are very much second-order influencers.
Similarly, all the cars are using the same tyres, which gives a degree of performance normalisation - especially on a single qualifying lap on new tyres. While it's true that the current tyres are very sensitive to operating conditions, when analysed statistically over a number of events, a pattern should rapidly emerge.
This leaves aerodynamics as the most likely cause of disparity between expected and actual results. But surely the sophistication of the tools these days should put paid to that type of concern? Oh, that it were so.
The aerodynamics of an open-wheel racing car are immensely complex. Much of the flow is turbulent and this makes it more difficult to predict using Computational Fluid Dynamics (CFD), and even a windtunnel, since there are many aspects of the flow that are difficult to model.
Both CFD (in the form commonly used by teams) and windtunnel testing operate in a time-averaged domain. What this means is they effectively measure forces or conditions which will be varying over a period of time, and then take an average result over this time period. In isolation this can give a very poor picture of actual performance.
Let's take a simple case where the downforce is varying in a simple, relatively slowly repeating pattern, like a sine wave with a variation that goes both 10% above the average and 10% below. The aerodynamicist will report the average number but the driver will only be able to exploit something around the minimum downforce, which is maybe 8-9% below the average.
Now consider a car that has nominally the same downforce, in other words the average figure is the same, but now the variation is plus or minus 15% from that average.
The driver once again will drive to somewhere just above the minimum, say 13-14%, which will give them considerably less downforce than on the previous car, even though the reported number is the same. In fact it may be worse than this because the second car, having more variation, will undoubtably give the driver a feeling of severe instability, discouraging them from finding the limit.
In addition to these sort of problems, it is common to hear teams reporting that they have lost the correlation between their windtunnel and the car on the track. Most people assume this means that the aerodynamic measurements made in the windtunnel are different to those measured on the car. There is an infernal triangle consisting of CFD results, windtunnel results and car results, and none of them give a true answer. One might think that the car must be the best measurement but unfortunately a racing car isn't a scientific device, and by instrument standards the measurements are somewhat crude.
On top of all that, the aerodynamic forces experienced by the wheel, be they the lift of the rotating wheel or the downforce produced by the brake duct winglets, are difficult to measure because they aren't passing through the sprung mass.
So what is good correlation? The answer lies not just in being able to measure numbers that when fed into a simulation will give a reasonable prediction of lap time, but more importantly a situation whereby trends can be followed with confidence. If the flow field predicted by CFD or even that experienced in the windtunnel is somewhat different to that seen on the car, then it becomes extremely difficult to predict the effect of changes.
When Williams had a problem after fitting a new rear wing at Silverstone this year the problem didn't lie in the rear wing itself, but in the flow detaching from the floor when the DRS was used and then not re-attaching when the DRS closed again. Not only was this not detected in the windtunnel, it also wasn't detected during practice the previous day.
This is a measure of just how critical the aerodynamic performance can be to the conditions under which it performs. When I was at Williams we even used to clean the leading edge of the rear wing in pitstops because the dead flies it collected could adversely affect its performance.
Good correlation exists when any trend seen on the car is predicted by the experimental techniques used to develop the car. Easily said but not easily achieved, and the more complex and critical the flow fields are that deliver performance, the more difficult that nirvana of correlation becomes.